Are actuators the bottleneck of humanoids or robotics? Which part exactly limits supply? Who are the main suppliers for each component that goes into the actuator? In today’s Part 3, we cover the global supplier map of actuators, for each component and where the facilities sit. We also look at where actuator supply could become constrained.

I spent the last month reading through dozens of corporate filings from the Chinese robotics supply clusters, and they sharpened the question for me. Scaling actuator supply means securing qualified parts for a particular robot design, at the required volume.

The supplier map shows where to investigate: precision manufacturing, motor performance, and materials and bought-in components. Those exposures move differently. Reading Laifual’s filings made me separate two questions: what a reducer costs, and how much qualified output a supplier can deliver. Today, we map the global supplier reality layer by layer.

Subscribe to get the next ones in your inbox. If you work in this space, please reach out. I would love to compare notes.

This piece covers six sections.

1. Bill of Materials: The Processes and Players Behind the Joint

2. The Global Actuator Footprint: East vs. West

3. The Friction Points: Feedstock vs. Process Control

4. Falling prices can coexist with tight capacity

5. Magnets: The Downstream Chemical Chokepoint

6. What could limit actuator supply at scale?

1. Bill of Materials: The Processes and Players Behind the Joint

We walked through each component from an engineering perspective, covering the metrics and what they do in Post 1, section 1. To dig into the supply chain dependencies, I find it helpful to look at these components from a manufacturing lens: the material, manufacturing process, and supplier set.

Quick refresher: the actuator is the muscle of the robot. It moves the joint. A shoulder actuator needs to be strong, since it is lifting a heavy arm and whatever the hand is carrying. A finger actuator’s speed requirement depends on the task.

Stator and motor. One option is a frameless brushless motor. In a slotted design, the stator is a stack of steel laminations with copper windings. The rotor carries permanent magnets. Frameless versions ship as rings that the robot maker fits into its own housing. Unitree’s filing states its motors are self-developed, and that it outsources some production steps, winding among them. Merchant suppliers include maxon, Kollmorgen (a Regal Rexnord brand), Novanta, MOONS’ and ThinGap.

Rotor magnets. One option is sintered neodymium-iron-boron (NdFeB), made by powder metallurgy: mill the alloy, press it in a magnetic field, sinter it, machine it, coat it, magnetize it. These are the magnets whose field interacts with the energized stator windings to produce torque: they are what makes the actuator spin. Processing is concentrated in China, which accounted for 94% of global sintered NdFeB magnet production in 2024, according to an IEA report published in April 2026. [1] Suppliers include JL MAG, Ningbo Yunsheng, Earth-Panda, Vacuumschmelze, Proterial, Shin-Etsu Chemical and TDK.

Reducer, low-ratio planetary. This is a quasi-direct-drive route, covered in Post 2. Precision planetary gearboxes can use ground gear teeth. Planetary-reducer suppliers include Zhongda Leader and Newstart.

Reducer, strain-wave. This is a precision route for robot joints. A strain-wave reducer uses a thin steel flexspline that deforms during operation. Suppliers include Harmonic Drive Systems, Nidec-Shimpo, Leaderdrive, Laifual and Schaeffler. Harmonic Drive LLC also has a Massachusetts production facility. This layer is expanded in sections 3 and 4.

Reducer, RV and cycloidal. This is the heavy-load precision route. The named suppliers are Nabtesco of Japan, Shuanghuan via its Huandong unit, and Zhongda Leader.

Planetary roller screw. This is one linear actuation route: a rotary motor turns a roller screw to create straight-line motion for high-force joints. It has a threaded screw, a set of grooved planetary rollers, and a nut. It is difficult to make at the high end because several rollers must share the load at once. Small errors in thread shape, pitch or alignment create uneven contact. See more in Section 3. Suppliers include Rollvis and GSA of Switzerland (one group since 2016), Ewellix (a Schaeffler company, with roller-screw production at Armada, Michigan), and Bosch Rexroth. Chinese entrants include Leaderdrive, Seenpin, and Hengli, which reported sampling and initial mass production in 2025.

Encoder. Encoder options include magnetic rings and optical discs read by sensors. Suppliers include Renishaw, Heidenhain and Broadcom.

Output bearing. Cross-roller bearings are one option for supporting joint loads. Suppliers include IKO, THK and Schaeffler. Laifual reports making crossed-roller bearings for its own products; Leaderdrive’s SKF joint venture is focused on R&D and industrialization of high-precision robot-joint bearings.

Driver board. The board carries power-switching and control electronics on a PCB. Servo-drive suppliers include Leadshine, Elmo and Kollmorgen; CubeMars lists a standalone driver board.

Housing and brake. The housing and brake depend on the joint design. Spring-applied brakes are one option, such as Mayr’s ROBA servostop.

2. The Global Actuator Footprint: East vs. West

The supplier examples span Japan, Europe, China, South Korea and the US. US production includes Harmonic Drive LLC’s Beverly, Massachusetts facility for precision reducers, Ewellix’s Armada, Michigan roller-screw plant, ThinGap’s Camarillo, California motor production, and eVAC’s Sumter, South Carolina magnet plant. VAC said in July 2026 that eVAC began commercial magnet production in 2025. Sanhua said in its 2025 annual report that it is expanding overseas production of its actuators, and updated in Aug 2026 that it’s processing through batch delivery and production-line ramp.

These are representative suppliers of actuator components. Exhibit 1 below records disclosed manufacturing locations where they could be matched to the product; other entries are marked manufacturing location unverified.

China’s actuator market is split between precision transmission specialists, integrated module suppliers, and automotive Tier 1 manufacturers pivoting into robotics.

3. The Friction Points: Feedstock vs. Process

Now this is the exciting part. Now that we know at a high level how each component is made and who makes it, it’s time to zoom into the requirements and potential exposures for each component. Broadly speaking, there are two places to investigate. (1) Feedstock and bought-in parts: the raw material, plus the merchant components that arrive finished, like sintered magnets and encoder chips. (2) Process and machining: what turns the raw material into the part, and the machines that do it.

Strain wave reducer. The process requirement is manufacturing control and accumulated know-how around the flexspline. The manufacturing challenge is making a thin flexspline that repeatedly deforms while maintaining gear accuracy and the required fatigue resistance. Material selection, heat treatment and machining must be controlled together.

Planetary reducer. The process requirement is manufacturing precision gears. Grinding the gear teeth is one process used in precision planetary gearboxes. Newstart, a Zibo, China-based maker that filed for a ChiNext IPO, reported ASPs for its mainstream 50-240 mm planetary reducers of RMB 780 in 2023 and RMB 791 in 2025. Its whole-company gross margin fell from 55.49% to 48.05% over the same period. [4]

Quasi-direct drive. The performance requirement sits in the motor itself. With less gear reduction, the motor must supply more torque for the same joint requirement. Take Direct Drive Tech as an example. This company sells direct-drive actuator modules for consumer, industrial and commercial applications. Its draft PHIP shows that, from 2023 to 2025, enameled copper wire rose from 6% to 27% of group material costs while the magnet share fell from 23% to 18%. These are group material-cost shares across its product mix. [5]

Planetary roller screw. The potential process exposure is the machining step. Leaderdrive’s H1 2026 report describes Chinese industry dependence on imported high-precision thread grinders, with high equipment prices and long delivery times. [6] Seenpin’s June 2026 application similarly describes reliance on imported high-end precision grinders. [7]

Rotor magnets. The potential exposure sits upstream in rare-earth separation and downstream in NdFeB sintering capacity, with China at 94% of global sintered NdFeB production in 2024, according to an IEA report published in April 2026. [1]

Encoders and driver boards. The potential exposure is bought-in availability: merchant chips and read-heads for the encoders, and the power semiconductor chain for the driver boards.

4. Falling prices can coexist with tight capacity

Laifual’s disclosures put price and capacity on the same page. They also show why a lower average selling price does not answer whether a supplier can meet the qualified weekly output for a particular design.

From 2023 to 2025, Laifual’s blended harmonic-reducer ASP fell 28%, from RMB 795 to RMB 571. Its small-size category fell 37%, from RMB 634 to RMB 397. These are category averages, not same-SKU price changes. Laifual attributed the 2023–2024 decrease to a higher share of small-size products and strategic price adjustments, and the 2024–2025 decrease primarily to the higher small-size share. Over the full 2023–2025 period, its harmonic-reducer sales rose from 115,315 to 291,515 units. [8]

H1 2026 tells a separate capacity story. Laifual reported designed harmonic-reducer capacity of 270,000 units and actual production of 256,900 units, a 95.1% utilization rate. The company also said that, against a backdrop of earlier relatively constrained production capacity, it had prioritized industry-leading customers. By July 2026, designed capacity had reached 64,800 units per month, so the H1 utilization figure should not be read as a September measure of spare capacity. [9]

A September 2 seller offer adds a point-in-time procurement reference, not proof of a fulfilled ramp. Laifual offered size 17, 20 and 25 component sets, each at a 100:1 ratio and a quantity of 100, for US$145, US$160 and US$173 per set EXW, respectively, with a stated lead time of three to four weeks. [10]

The buyer question is concrete: for the specified part and qualification standard, how many units can the supplier deliver each week while maintaining quality?

5. Magnets: The Downstream Chokepoint

Magnets represent a highly concentrated layer of the actuator supply chain. Recall that permanent-magnet motors carry magnets on or within the rotor. Electric current flows through the copper windings in the stator, creating an electromagnetic field. The rotor magnets follow that field to spin the motor. Torque per unit of mass is a critical metric for a humanoid robot that must carry its own weight.

These are made from a sintered alloy of neodymium, iron, and boron (NdFeB), commonly called “neo magnets”.

It turns out that “rare earths” are actually not that rare. Economically viable concentrations are uncommon, and their chemical similarities make them hard to separate. [11]

To map the potential exposure for robotics, I group the value chain into 3 phases using International Energy Agency (IEA) data. All three are concentrated in China:

1. Mining (Upstream). Hard-rock ore is extracted, crushed and concentrated. China supplied 60% of the four magnet rare earths (neodymium, praseodymium, dysprosium and terbium) in 2024. Other producers include Myanmar, Australia and the US. [1]

The potential exposure here is access to economically viable deposits. Extraction routes differ: hard-rock ores require crushing, while ion-adsorption clay deposits, including those in Myanmar and southern China, can be leached.

2. Separation and refining (Midstream). This is a highly capital-intensive and chemically complex stage, and China accounted for 91% of refined output of the four magnet rare earths in 2024. A common process is Liquid-Liquid Solvent Extraction, routing dissolved rare earths through repeated extraction, scrubbing and stripping stages. [1]

The potential exposure here is environmental policy and process know-how. Waste treatment depends on the ore and processing route. Where the ore contains thorium or uranium, processing can concentrate radioactive material in residues.

(3) Sintered permanent-magnet production (Downstream). Metals are alloyed, milled into powder several microns in size, pressed in a magnetic field and sintered. [12] China accounted for 94% of global sintered permanent-magnet production in 2024. [1]

Two processing challenges in this stage are machining yield and grain boundary diffusion.

• Machining and Yield Loss: Shin-Etsu describes grinding sintered magnets with diamond abrasives to reach the required dimensions. Removing material during machining creates a yield-management problem. [13]

• Grain Boundary Diffusion (GBD): A heavy-rare-earth diffusion source is brought into contact with the magnet, followed by diffusion annealing. This can improve coercivity, the magnet’s resistance to demagnetization. [14]

Getting magnet supply set up is so much more than finding and opening a mine. The potential constraints include navigating midstream separation chemistry, waste management costs, and downstream precision processing.

§6: What could limit actuator supply at scale?

Congratulations on making it to the end of this post! This piece has got longer than I originally anticipated. But this also shows how deep and complex supply chains are. As a final reward of finishing the post, let’s apply what we’ve learned on the supply chain to tackle the million dollar question: if annual humanoid production rises from an illustrative 20,000 robots to one million, what breaks first?

Read more

Scott is a mechanical and aerospace engineer, a 2-time robotics founder, and one of the sharpest public readers of humanoid-hand architecture. He has worked through Tesla’s hand iterations, Figure’s decision to abandon its first tendon-driven design, 1X’s routing choices, and the direct-drive approaches behind Wuji and Sharpa Wave.

This was a particularly fun and wide-ranging conversation. We used three extreme designs to escape the usual tendon-versus-motor argument: Allonic’s braided structure, Daxo’s maximalist system with up to 120 tendons, and Tacta’s hydraulic hand, glove, sensing, and data stack. Tacta is especially interesting because it is designed as a complete manipulation system for cobots and industrial arms, without trying to fit inside a humanoid.

The side paths were just as useful. Scott proposed a decathlon for humanoids, complete with pole vault, hurdles, baton exchange, and the same sand pit humans use. We also asked whether robots should copy human movement at all, and whether a general-purpose body still becomes specialized through its profession, as humans do.

In this episode we cover:

• Why Tesla and Figure reached different conclusions about tendons

• What Bowden tubes solve, and what they add at the wrist

• Why joint count and independent control are different metrics

• What the Humanoid Games reveal about robot-specific movement

• Scott’s proposal for a humanoid decathlon

• What Allonic, Daxo, and Tacta teach at the design extremes

• Why Tacta is building a manipulation system outside the humanoid

• How generalized bodies can still become specialized workers

Scott Walter: https://www.linkedin.com/in/scott-walter-ph-d-b2a78ab/

RoboStrategy: https://robostrategy.co/

Watch on YouTube: https://youtu.be/VbwM0t1oJQI

Every episode, I sit down with the founders, researchers, and operators building physical AI worldwide, tracing the full stack from components and supply chains to what actually deploys. See all published episodes here.

Chapters

00:00 Cold open: the labor prize and the 5-year forecast

00:24 The Fermi paradox of robotic hands

01:18 Tesla, Figure and the tendon debate

05:10 Bowden tubes and routing tendons through a wrist

11:15 Joints, actuators and real degrees of control

14:38 Why the pinky matters more than it looks

17:23 Humanoid Games as real-world robotics experiments

21:57 Scott’s humanoid decathlon challenge

26:28 Tesla’s hand iterations and abandoned designs

27:55 Wuji and Sharpa Wave put motors in the fingers

38:29 What 3 extreme hand designs can teach us

38:53 Allonic’s braided hand and portfolio disclosure

51:15 Daxo’s maximalist hand with up to 120 tendons

01:02:36 Tacta measures how workers use their fingers

01:05:18 A robotic-hand system designed outside the humanoid

01:13:12 Data-center cables as a tactile manipulation task

01:16:22 General-purpose bodies and specialized work

01:19:18 Why this 5-year robotics forecast may be different

Full Transcript

Below is the full transcript. There may be errors as it’s AI-transcribed.

00:00:00 — Cold open — excerpt montage

Scott Walter · 00:00:00

That is the biggest pie we’ve ever seen, ever, trying to automate labor. And sim-to-real gap is almost zero. So locomotion is a solved problem. Humanoids will be deployed like five years after everyone stops laughing. And they’re the worst they’re ever going to be. They’re just going to get better and better and better.

00:00:23 — Why dexterous robot hands are still hard

Scott Walter · 00:00:23

And what I call the Fermi paradox of robotic hands. If they exist, where are they? And that’s always been the problem: you see them in the labs and then you go around and you’re like: I’m not seeing them anywhere. No one’s able to do them. And the question is: why is that the case? It’s extremely challenging to come up with something that has not only dexterity, something you can control, but also robustness. But what I was seeing during the teens is that robotic hands were suddenly becoming a possibility because machine learning made it possible. Before then, they were just really, really hard to control.

Scott Walter · 00:00:56

And that’s what the Shadow Hand proved: you can have this tendon-driven hand. And if you connect it to really good machine learning algorithms, you’d be amazed that you’d be able to do. The problem was that the robustness wasn’t there, the cost factor, everything else. So that’s kind of the preload that we’re seeing it. And why aren’t they becoming more and more available? Now, the big debate that is going on in hands in general.

00:01:18 — Tesla, Figure, and the tendon debate

Scott Walter · 00:01:18

And so people really started paying attention to robotic hands when Elon announced a 22 DOF hand that was going to go on Optimus and the revealing of that and everything like that. And after that it was like this Cambrian explosion of just robotic hand designs all over the place and everyone pursuing different designs. And for the most part, you could break it down into two categories, tendon-based hands and non-tendon-based hands, which most people would say direct drive. But even there, you have to break them down because you can never really put things in two different buckets.

Scott Walter · 00:01:52

They’re all going to go down there. But you can kind of look at it that way. And everyone’s pursuing it. And they all were very passionate about their designs and why theirs was the best approach. Some people would try one and they would kind of think about it again and then try a different approach. But about two weeks ago, Brett Adcock put out a post where he talked about the evolution of the hand designs at Figure and that the very first generation hand that they built was a tendon-based hand. And after attempting to do that he came to the conclusion that a tendon-based hand was basically an engineering dead end.

Scott Walter · 00:02:24

And he called it the biggest engineering mistake he’d ever made. And since then, he’s been pursuing hand designs that are non-tendon-based. And that kind of caused, let’s say, a very spirited discussion. And I’m glad it was out there because you got a lot of people that were very much on the side of tendons just going out and just saying he’s completely wrong. From right down to saying it’s a skills issue to explaining why they believe it and their particular thesis for it. As well as you get a lot of support from other people who said, yeah, we tried tendons as well.

Scott Walter · 00:02:55

It turned out to be really hard. Now, I’m not here to defend one position or not. I’m really here to maybe kind of present it to everyone so everyone can kind of come to their own conclusions because they both have their pros and cons. They both have the big engineering challenges in what they’re trying to do. And even within the tendon community, while they agree tendons the right approach, they will get in very spirited arguments about what’s the best approach or not. And you can see that 1X came out with their 22 DOF hand, I think.

Michelle Sun · 00:03:25

I think it’s 25.

Scott Walter · 00:03:28

Yeah, 20. I think it’s actually 22, 44 tendons. It’s kind of hard to know the count. They’re also throwing the wrist in there. So I usually don’t put the wrist in there. But they have what you would typically say is about 22 degrees of freedom of movement of the fingers. Quite complicated design.

00:03:43 — Routing tendons through the wrist

Scott Walter · 00:03:43

And the biggest challenge with all the hands is not having the tendons that can actually move the fingers and putting down the forearm, but figuring out how to get them through a wrist. The wrist isn’t in there. It’s an easier problem. I’m not going to say it’s easy. It’s not. But as soon as you do that you add this extra level of complexity that the engineering solutions are challenging. And bodies are fairly simple to model. But when we start getting into flexible stuff, there’s a lot of challenges. And we are doing it in our heads because we’ve been doing it for so many times that we know we can move our wrists and somehow keep our fingers constant.

Scott Walter · 00:04:19

But if we’re not doing that we will see that our fingers will kind of move as we are moving our wrist just because the length is changing. Now, in theory, you can solve the problem by having all your tendons go right through the center of rotation of your wrist. Okay? In theory, you could do that. The reality is there’s something called, I think it’s the Pauli principle. Two particles cannot occupy the same point in space at the same time. So you need room. It’s like you’re talking about before tendons. They can’t all occupy the same spaces.

Scott Walter · 00:04:51

So when they go around there, there’s just always going to be that. And then there’s this other problem is that when you’re doing it, sometimes if the tendon’s coming straight up and you’re going, that means suddenly you have to do this very sharp angled turn. And tendons do not like sharp angled turns. You need to have a kind of radius in there. So there’s big challenges to being able to do that. And there’s different schools of thoughts on how to do it. And there’s this thing called a Bowden tube. And everyone’s like: whoa, what’s that?

Michelle Sun · 00:05:15

Is that what 1X is?

Scott Walter · 00:05:20

Yeah, they’re using the Bowden tubes. And basically, if you’ve ever ridden the bicycle and look at your brakes and the brake cables, that outer tube on the outside, on the inside is a wire, which is effectively like a tendon. And the tube helps redirect that and maintain the same length. That’s the trick. That’s why you can turn your handlebars without suddenly the brakes engaging or disengaging or something like that. Or your gears changing or anything like that because you’re able to maintain the length through those things. Now, those are already engineering challenges. They have friction.

Scott Walter · 00:05:53

They take up space. They can snap as well. So within the hand community itself, there’s like a big debate on whether to do Bowden tubes or not. And the original Tesla hand that they showed at the 1010 event had them. And then the next time they had the public showing of them, they didn’t have them. And so everyone’s trying to figure out how do we route these things through there. And this is why you can see all the people that are building the direct drive ones have given up. They just say that is just a challenge.

00:06:23 — Tendon materials, flexion, and extension

Scott Walter · 00:06:23

The other main thing is what is the material you use for a tendon? If you use a cable like you see in a bicycle, the problem with that is the bending radius is not very good. And they tend to elongate over time. Anyone that’s had a bicycle knows you have to go and retune everything and make those adjustments because they’re going to stretch. And the other thing about metal is that anytime you bend it, there’s energy involved in the flexing of it, which means that’s energy going not to your hands. If you use something like rope or thread, it has much better properties that way.

Scott Walter · 00:07:00

It’s very flexible, very easy to move around. The radius of curvature is a little bit different, but they can fray and also stretch and everything else. So you’re worried about wear and tear. But there’s this magic material out there called Dyneema, which is used for climbing ropes and everything else is extremely flexible. It’s this aramid polymer, which is just kind of miraculous and doesn’t stretch very well and has very good abrasive properties. And it’s readily available. You can go on Amazon. You can just order it in any color you want. So anyone that is, an amateur building their own hands at home by using a 3D printer, and there’s plenty of open source hand projects out there, the Paloma project is one example.

Scott Walter · 00:07:39

You just put that in there. So it’s very easy. And they’re all using the same thing. Rock climbers use it for a reason because it’s like it doesn’t break and it doesn’t stretch. And also, it’s got this one great property that another kind of aramid known as Vectran doesn’t have, and that is it does not break down with UV light. Okay? So Vectran is very sensitive to UV. It will break down. But it has some properties a little bit better than Dyneema. And if it’s inside, ah, maybe it is that. So usually it’s two of those.

Scott Walter · 00:08:10

Or there might be some sort of, magic hybrids of these things where sometimes people might even put in stiffness. So by adding metal. One advantage of a wire cable: the problem with any sort of thread is that it’s really great in tension, and as soon as they’re trying to push on, it collapses. So I can only apply force in one direction. Bicycle cables, we know, work better in one direction, but they still work a little bit in the other. They will a little bit, but usually on the brakes, we still have a spring to help open them up.

Scott Walter · 00:08:42

The problem with a tendon-based hand is that you’ve got two ways of moving. One tendon gives you the flexion. And in order for you to open up your grasp, what they call extension, you need a second tendon, which is called extensor, to pull it back open. Or you need a spring return. And that’s not necessarily a great way to do it. And that’s because when you push on a thread, you’re not getting anything out of it. With a Bowden tube, you can do it a little bit, but mainly all you’re doing is taking the pressure off as opposed to pushing it back.

Scott Walter · 00:09:15

Unless it’s a bicycle cable and unless it’s human tendon. It turns out human tendons are an amazing thing because they can take pretty heavy loads and they actually can give a little bit. So our tendons are able to push back, not 100%, but much better than Dyneema. So there’s already a difference between what they’re using and an actual human tendon. Yeah. No one’s made that material yet because it has, the properties of a metal cable without being a metal cable. It has the properties of a Dyneema without being a metal cable. It’s like if someone could figure out how to make that thing, that would be a lot better because then you would be able to push pull.

00:09:50 — Direct drive, linkages, and degrees of control

Scott Walter · 00:09:50

And this is where the people that want to have what I call the direct drive are going to be in there. And there’s a couple ways to break down the direct drive. The first way is, everyone was trying to say, where’s the space to put our motors? If we can’t put them down in the forearm because we don’t want to have to go through this wrist mechanism. Well, it seems like there’s a lot of space in the palm. So everyone says, let’s go ahead and put something there. And the first generation Tesla hand did that but they actuated the hand actually with a tendon, which is a metal tendon.

Scott Walter · 00:10:20

And they realized all those things stretch and stuff like that. But the run link was very short and you didn’t have to run it through here. Others said, well, wait a minute. There’s another way that we can do that have the same thing with a tendon. And that is to use a linkage-based mechanism for that. So they created basically these mechanical links that go through there. And the advantage of a mechanical link is that it works in both directions, in tension and compression. So that means you only need one motor and it can give the same amount of grip strength as well as opening strength.

Scott Walter · 00:10:50

It can go both ways and it can lock in really tight and everything like that. Whereas if you’re using a tendon, you would need the second tendon or you need a spring return. In some cases, you might need a second motor unless you wrap it around like a capstan. But then it’s hard to separate the tugging from both of them because there’s a lot of complications here. Now, when we talk about degrees of freedom in joints, they’re two completely separate things. We have a lot of joints in our fingers. Our fingers have, you consider, four degrees of freedom.

Scott Walter · 00:11:21

Now, you might say there’s only really four IP joints. One of them is not called an IP joint. It has to be called the MCP down here. But it has two degrees of freedom. It can flex, and it can do this ab/adduction that everyone talks about. And then we have the other IP joints here. But for the most part, there are four ways that it can move, four joints there. The question is, how do you control that? And degrees of freedom really means more or less degrees of control. So if I have a single tendon through there with a single motor, I have only one degree of freedom.

Scott Walter · 00:11:49

And then what’s going to happen is there’s going to be coupling between them and my finger is just going to move like that. It’s very difficult to get different degrees of control without adding more motors for that. And so what ends up happening with a linkage-based mechanism is that you already have the coupling. And in many cases, there’s only four or five motors in there. So those fingers really only had five or six degrees of freedom, depending by the number of motors, even though they may have had like 15 joints, 15 ways of moving. And so what happened is that they would collapse down like that.

Scott Walter · 00:12:21

We assume they’re all identical. It’s like: nah, there’s a lot of stuff. So the thumb has way more kind of range, different kind of range of motion, more tendons than the other. I think our thumb has anywhere between eight and nine. And it’s like: well, what’s the answer? And it’s like: depends who you ask because it just shows how kind of nuanced some of these answers people are going to have. But typically, it’s more than what you would see in the others. And the thing with this closing like that is that we really don’t want our fingers to close this way where everything is coupled and there’s only one way of closing.

Scott Walter · 00:12:51

And that is, it means when you go to grab a bottle or something like that your fingers can actually miss. Try coming down and suddenly, we want to have the straight move and then have that clasping. So that’s something that’s important. Now, you can kind of do that with a mechanical linkage in that what you do is you have like an escapement mechanism, block mechanism. So it’s like the very clever that you can get it that when it’ll move like this and once it gets there, something kind of engages and now the fingers do this.

Scott Walter · 00:13:17

Okay. So you get this nice illusion that you have more degrees of freedom of control there and you really don’t. You’ve programmed it in. It means, oh, I can’t do this. I have to do that to close it. It’s the reality. But it’ll work in a lot of cases and you can get away with fewer motors and you do have that nice strength of being able to open things up really quickly. So that’s a lot of the advantages of this direct drive: it’s almost tendon-based, except the tendons are super rigid. One of the problems is when you get those mechanisms, it’s pretty easy to build linkages that will just give you this.

Scott Walter · 00:13:49

But once you do that you’re starting to have that same problem we talked about with the wrist. Okay, I get some mechanism that’s pulling up and down here. How am I going to build this linkage mechanism that not only is going to do this, but allows me to suddenly redirect everything along there. It can be done. It’s just that it’s a more complicated mechanical design and then the decisions of how you control the abduction. So we’ve got basically five ways of doing abduction. Do we have five motors or do we have a single motor? Those are the decisions everyone has to make.

00:14:18 — Why the pinky matters

Michelle Sun · 00:14:18

Yeah. And one of the things that stood out to me about the fingers is that for carrying heavy loads, right, I read that the pinky actually does a lot of work as well. So it’s more than we think and we don’t think about it when we do that.

Scott Walter · 00:14:35

Oh, way more than you think. The other thing that pinky does: there’s this big argument, and one of my podcast colleagues, Gustav Andersson, has pointed this out. I’ll ask you a question. If you had to lose one of your fingers, which one would you choose?

Michelle Sun · 00:14:54

Probably the ring finger.

Scott Walter · 00:14:56

Okay. Some people say the ring finger. Some people would go with a pinky. And it turns out he actually says it’s the index finger. I’m like: what? He’s a hand surgeon. He knows better. And part of it is because if you take out the ring finger or something like that you get like a weird gap. The pinky is more important than we think. And it turns out that our middle finger can take over for the index finger very easily. People are able to get that kind. That was, kind of shocking to me. And I only assumed it.

Scott Walter · 00:15:22

And the main thing about the pinky is not so much that it’s strong, but where it is, where it’s placed. When you want to grab something with authority, if I’m just grabbing it with these two fingers, you see it wants to move around. And if I get the others in there, it still can move around a little bit. And you’ll see this when, these robot hands are picking up a hammer. A lot of times, the hammer is really not in there. Well, it turns out when you get that pinky engaged in there, suddenly that thing is, just mechanically locked better than when the others are in there.

Scott Walter · 00:15:48

And a lot of it just comes down to simple physics, and that is you have kind of a fulcrum that whatever you’re grabbing usually is resting right in there. And then these others are trying to come around, but they don’t have as good a lever arm, right, because the distance of where that’s happening. The pinky comes down here, and the distance from here to here is so big. It doesn’t have to be strong. It gets it just from simple mechanical advantage of the distance.

Michelle Sun · 00:16:10

Yeah, it’s locking in that grip and stabilizing it.

Scott Walter · 00:16:14

And you see that with a lot of tools that you have, the way the grips are all set up. Is that pinky is in there that gives you that final amount of stabilization that is remarkable. And again, it’s just kind of a geometric thing. And why the nuance and the subtlety of what goes on with hands is so underappreciated until you start going in there. Now, this is the other thing that I invite everyone at home to do. A lot of times when you’re designing stuff, you have to make these prototypes and experiments. So, if you’re building a car or drones, you just got to go in there.

Scott Walter · 00:16:45

You can think about it a little bit, but you got to build the darn thing. When we start making hypotheses about how a humanoid works, you don’t have to build a humanoid. You are a humanoid. You can run the experiments on your own. And that means you can ask the question, what happens if I reduce the degree of freedom? What if this is removed? What if I try to do everything in the house without my pinky? So, just attempt to see what you can do and how important it is. Take, basically, medical tape, something like this, and just, bind some of your hands.

Scott Walter · 00:17:15

Or put on, really thick gloves. Restrict motion of, one of your shoulder joints. And just ask the question.

00:17:23 — Humanoid Games and real-world experiments

Scott Walter · 00:17:23

And so, a little bit of a cul-de-sac here: the humanoid games in China were fascinating to me. Because we get to see people running these real-world experiments. So, they had these highly optimized bots for running in the 100. And we could say, they’re not truly humanoid because they took away a lot of degrees of freedom. But I am so happy they did that because we are always getting in an argument about how important is that joint to running? How many degrees of freedom do we really need in the shoulder to be able to run effectively and stuff like that?

Scott Walter · 00:17:55

And you can run them in a simulation, and they kind of give you an answer, and you can intuit yourself. But there’s nothing like real-world data. And seeing how they did it, and then you begin to go, huh. And you might say, yeah, what’s important is, we finally understand the nature of a particular joint or particular axis to a degree that we’ve never had before. That we now can understand what’s a better way to optimize it when we want to build the complete one. So, I love the fact that they’re going through, and you have these very specialized events that are forcing you to answer that question in a way that can only be demonstrated with a real running experiment, if you want it that way.

Michelle Sun · 00:18:35

Yeah, for sure. And one of the races I saw that some run-up to humanoids were, leaning forward the whole way. And that was just, really fascinating and, different posture that usually we don’t think about running in, and they just experimented with it.

Scott Walter · 00:18:52

Yeah. And I think one of the best ones is, you may have seen it, is, everyone complains about, AI slop videos or, how people just use AI in a way that’s oh, it’s so bad. And you may have seen someone actually took one of those races and replaced them with humans. So, you saw the human runners, and I was this is great, because what more effective way of explaining what that running form looks like than to do that direct comparison of how they were leaning and everything else? The other thing is, we’ve got to remember, is, they are not built exactly like us.

Scott Walter · 00:19:26

They may look like us, but their musculature is very different. So, they have different power density. They have slightly different kinematics. The weight distribution is very different. So, we shouldn’t be surprised that they have this awkward running form. Now, I do not recommend that human runners run with their arms like that and lean forward, because you’re going to end up with, back pain. But it turned out that the engineers were trying to solve a particular problem, and it was, kind of overheating. And they ran, RL sims to see how we could do that to minimize the movement on one of the shoulder joints.

Scott Walter · 00:19:57

And that was the solution it came up with. And it turned out the solution was highly optimized, better than they all expected. And this is the problem with dealing with human priors, is that everyone’s, forcing their walking policies to look human-like. But it may be that a human-like walking or running policy is not actually the optimum. It might be the suboptimum for these creatures. And so, they come up with it. And I think someone mentioned it. It’s basically, it evolved the tail backwards. Because, velociraptors and others, they need the big tail back there to do that.

Scott Walter · 00:20:25

And effectively, they said they weren’t allowed to have a tail, but they kind of did it that way. It’s, yeah, that’s an interesting way of thinking the problem. So, as silly as that looks, again, for me, I am super excited because I’ve always wanted to know. These are these academic debates that we have all the time. No longer academic. We can point to something. And this is why he’s saying these races are going to be studied for years. Even though they’ll be surpassed by all sorts of things, when you go into teaching any sort of engineering class, you’re going to want to bring that thing up and just say, take a look at this.

Scott Walter · 00:20:57

Just like the Tacoma Narrows bridge disaster is still taught in every engineering class on structures and dynamics and everything else because it is so educational. And that 100 meter we saw, it was incredible because of the different phase transitions, again, pitter-patter, how they were going to accelerate, stand up, lean forward. What are you supposed to do? It’s, every track coach will love that as an example to show their sprinters, oh, this is why we want you to do this. Do you understand what we’re talking about now?

Michelle Sun · 00:21:29

Yeah, and I love that they actually have different games, just like the Olympics, right? There’s the sprints, there’s also the long jumps. So the makers can really, design very different robots to optimize for that and train different policies. So it really is pushing to the extreme: what is possible? And how do we rethink the whole architecture from, the joints where they’re put and how they run, how they jump?

00:21:57 — The humanoid decathlon challenge

Scott Walter · 00:21:57

Now, I pointed out already in a couple other podcasts, I’m going to do it again, and I’m putting it out there very, very soon, is I’m throwing down the gauntlet for the humanoid decathlon challenge. And that is actually a humanoid that can do the decathlon. And that’s very challenging because it’s not just about locomotion anymore. There’s a certain amount of manipulation. We can argue whether the degree of the manipulation is totally at the level of human dexterity. And I think you’ll be surprised. It’s pretty close. So, arm strength is going to be incredible, whole body coordination.

Scott Walter · 00:22:30

You’re going to need really good grip strength. You’re going to need, basically, some wrist motions to be able to do some of the events. So, I think it’s a really pretty good proxy of coming up with a humanoid, which is going to be much closer to a real human. And then we can start arguing about world records at that point. So, we would say, what is the form that’s able to do it? And if it’s able to do it successfully, if it breaks the record, then we can say that’s on par. Because Usain Bolt is able to do a lot more things than the robots that broke his record.

Scott Walter · 00:23:02

And one of them, he was able to stop without running into a wall. Okay. He was able to run the curve. And the other thing he pointed out is, not only did he get a gold in the 100, he got a gold in the 200, which meant he had to run the curve. And those bots really could not run the curve because they reduced the degrees of freedom to make sure they’re really good going straight. And the last thing is the last gold medal he got in was in the 4x100 relay, which means there’s a baton exchange, which means there’s a certain amount of manipulation.

Scott Walter · 00:23:30

You actually have to be able to receive the baton and hold on to the baton. And so, in order to get parity, to make it fair, we have to say: what do we do to force someone to build a humanoid that has all these required degrees of freedom, that has to do the 100, has to do the hurdles, has to do the high jump? Has to do the pole vault. The pole vault is going to be a test of strength, not just grip strength, but upper body strength. We didn’t see any robots doing pull-ups. A lot of them aren’t even strong enough to do a pull-up.

Scott Walter · 00:24:04

It’s like we shame how a lot of people can’t do a one pull-up, but the humanoids really, they don’t have enough strength to be able to lift that. So, when you start thinking of an athletic competition like that’s going to be the true test. And again, I’m going to put out the ground rules and everything like that. I hope next year at the games that they have a decathlon in there and they attempt it, and it may be that they struggle. Just like last year, they struggled in the games, right? It was comical. This year, it wasn’t comical.

Scott Walter · 00:24:33

It was like Formula One, NASCAR, kind of disasters going in there, but at the same time, incredible performances. So, you can see where you can go in one year. One year, I have a feeling these bots might struggle through a decathlon, but then they will learn so much that it’s possible. It’s possible in two years they could make an attempt at the men’s world record in the decathlon. So, that’s out there. That’s the challenge I want to have. Let these labs get started on that.

Michelle Sun · 00:25:07

Yeah, definitely. Well, then you can start thinking about building that. And I think we can really expand into all the Olympic sports, right? Like the team sports can be expanded into, and like that will be really fascinating.

Scott Walter · 00:25:22

Cycling, kayaking. You can go down and down the list. And eventually, there will be a humanoid that will probably be able to perform in all of those things. But kind of steps. There’s a lot of things that – strength and dexterity in the decathlon that most people agree is pretty challenging. But there’s not going to be the finesse like we’re not going to have it play the piano when it’s over. Maybe we should. It has to play the Olympic theme music. So, some things like that we know are maybe not fully encompassing, but we need some sort of proxy.

Scott Walter · 00:25:54

We already have the human data on decathlons and how challenging it is. So, let’s start with something we’re all familiar with. And again, exactly the same one. Not oh, we’ll do an event kind of like it. It’s like none of it’s the same. Right down to the long jump. You may notice the long jump, they were landing on the surface. The real long jump, you’re landing in a sand pit. And I have a feeling they didn’t want to land in a sand pit for a reason. And that is, sand gets in your joints.

Scott Walter · 00:26:19

And so, it’s like: no, no, no, no. If we’re going to do this, we’re going to do it with the exact same little ground rules.

00:26:27 — Tesla’s hand iterations

Michelle Sun · 00:26:27

Yeah. And do you want to go into the Tesla hand still?

Scott Walter · 00:26:33

Yeah, we can briefly talk about it. Because, again, we don’t necessarily want to talk about the hands that have already been talked about to death. So, the lead up of the Tesla hand is that it’s doing it very much like a human. And Elon talked about the advantages putting everything down the forearm. And there’s a lot of people that believe it. And now you’ve got to put tendons in there. And the idea is that you want full hand dexterity. You don’t want these hands that has minimal amount of degrees of freedom. But full hand degrees of freedom so you can play the piano.

Scott Walter · 00:27:02

So, the first generation one was seen at the Tevent. And it was like the prototype. And that was showing Bowden tubes. And if you look at it, you will see that as the wrist moves around that these tubes are kind of bending and going all over the place. And you’ll see the same thing on the 1X hand that they have that. They then abandoned that and came up with something else. And this is like their next generation hand. And there was a lot of patents that were filed on it. There were four or five patents.

Scott Walter · 00:27:27

And then when the patents came out, I did a patent review with Humanoid Hub on that. And one replied to Humanoid Hub’s post on that. It’s like: oh, we abandoned that design. Something like that. It didn’t work. So, a lot of the things that they were working on turned out to be kind of dead ends. And they’ve had to do something else. Of course, we don’t know what that is. And everyone else that’s building tendon-based hands is discovering the same kind of challenges and problems.

00:27:52 — Wuji, Sharpa, and motors in the fingers

Scott Walter · 00:27:52

And then if we go over to the other side of people who are in the direct drives, There’s two other ways of doing it. So, the Wuji hand. So, normally everyone says, let’s put the motors down here and either use linkages or maybe short tendons to be able to do that control. Wuji said, all right, what we’re going to do is we know how to make really small motors. And we’re trying to find space for where to put the motors. And everyone wants to put it in the palm. The problem is you do that. The palm gets pretty big.

Scott Walter · 00:28:18

And they said, well, wait a minute. What’s sort of the biggest point of space in the finger? And it turns out it’s what we call the phalanx, or everyone would call it like the bone. And so, they said, why don’t we make the bones out of motors? Because they’re really big. The other alternative, which we see in like the Sharpa hand and origami and some others, is that to put it actually in the joint. So, build your finger like you would build a robot where every single joint is actually the actuator. The thing is, you don’t have that much space there.

Scott Walter · 00:28:48

And they’re looking at, how can we get something a bit bigger to get the torque density that we’d like to have? So, it makes sense to put there. There’s just one challenge with that. And like with everything, every time you come up with a solution, there’s like these other challenges you’ve got to get around. It’s kind of frustrating. By putting it right at the joint, the motor is spinning in the direction you want the joint to move. And that’s great. And you may or may not need a reducer depending upon your torque density. And what I mean by reducer: a reducer reduces the speed or something.

Scott Walter · 00:29:15

So, if something is spinning really fast and you don’t want it running really fast, you put a gearbox on there. With the advantage, not only does it slow it down, but it gives it more torque. And so, if you have a motor that’s spinning super fast and your finger might close, it’s like: no, no, I don’t want that. I want to slow it down. So, the reducer is there and also give you a bit more grip strength. And the reducer takes up a little space and everything like that. But that’s one way of doing it.

Scott Walter · 00:29:36

But with the Wuji hand, now we’ve got way more space. The problem is the spinning is along the axis of your finger and not perpendicular to it. So, you have to come up with some sort of transmission system to do that. There’s two ways. One is like a bevel gear design that anyone that’s kind of familiar with the differential in the back of their cars is that kind of design. The other is something called a worm gear. And I made a guess just based on kind of like the offset of where the actual joint is that they probably are using a worm gear.

Scott Walter · 00:30:07

And they are. So, the worm gear and the knuckle, they don’t necessarily show up. But I think there were a couple others that they were showing. So that works great for your IP joints. And the thing that’s nice is the distal, the very tip, your fingertip, they don’t have a motor there. Because the motor for moving it is in the one before, which is great. And then, the motor for this one is down here. And then, they get down to the MCP. And the MCP is always the biggest challenge because it has to move two directions.

Scott Walter · 00:30:36

Now, typically, the order of operations of those joints is that flex. And then, the abduction is always a joint before mechanically when they build it. Wuji decided to switch that up. And the reason they decided to switch it up is rather interesting. It might be partly because they figured that there’s an architecturally easier way to do it. And the other is the most important joint when it comes to like grabbing something, as you’re referring, like lifting up your luggage, is that joint. That’s where the grip strength is for trying to hold something, whether it’s a bag of groceries or anything like that.

Scott Walter · 00:31:06

And the other two, basically, they don’t provide very much except they make sure it doesn’t slide off your fingers. And so, that means you would like that thing to have a lot of power. And if you change the order of operations, then it becomes a bit more difficult. And they decided, we’re going to actually put a motor with three times the torque output as the ones that are up here. And we’re going to use that for flexion on the first joint. And so, they did that. And you’ll kind of notice as you look at that finger.

Michelle Sun · 00:31:33

Is that the circle, like the silver?

Scott Walter · 00:31:36

Yeah, you see that circle? That’s your abduction. That’s your abduction coming after the flexion. Right. And they had to come up with an interesting four-bar mechanism to be able to do that. But when I looked at it, they really understand the nature of the problem. And it’s like different than what everyone else did. Now, worm gears are really good at mechanical leverage, way more than most people would give them credit to. The problem is they’re so good at it, they’re very hard to backdrive. They work really well in one direction, but not so much in the other.

Scott Walter · 00:32:05

And you want torque transparency, and you want compliance and everything else. They’ve come up with a second version of a worm gear. There’s another type of modified worm gear that does give you backdrivability. But what it does is it reduces the forward advantage a little bit. It’s like everything. There’s always this compromise. But now they do have a little bit of that mechanical compliance in there. And that’s very important with hands and everything is we want backdrivability. We want them to be compliant. The biggest problem with a lot of these direct drive in the fingers is that you’re moving a ton of mass out into your fingers.

Scott Walter · 00:32:42

So the more mass you have in the fingers, the harder to accelerate, the harder to have compliance and everything else in control. And this is the argument that everyone that’s doing tendon-based hands, and especially like Kyber, which has a very good tendon-based hand, is that their fingers are just so light that they can move around really quickly. So, again, that’s the pros and cons. The other thing is that as you start scaling these motors down, unfortunately, the scaling problem doesn’t work as well as you would like. The motor will overheat very quickly. So not only do you lose mechanical advantage, they can get very warm very quickly.

Scott Walter · 00:33:16

And that’s a challenge. So the difference between Wuji, which is direct hand, and Sharpa, which is direct hand: Sharpa is going straight to the joints. And Wuji is doing it there. They still are putting some down here in the palm, but they have such good control of the fingers and the positions of them that their sim-to-real gap is almost zero. So every research lab wants to use them because the sim-to-real gap is gone, whereas with tendon-based hands, there’s still a lot of finagling. It’s still very challenging to do that. And that’s let’s say, their major advantage.

Scott Walter · 00:33:48

And then everyone will point out like all these other advantages. And that’s why you have these different camps that are arguing at each other of I can solve this problem. And then they’ll point out, yeah, but you can’t solve that one. And you don’t have the perfect hand yet as a result.

00:34:00 — Counting degrees of freedom and scaling hands

Michelle Sun · 00:34:00

Yeah, and Sharpa is the one that even the pinky has an extra degree of freedom, right? Is that what you’re just doing?

Scott Walter · 00:34:09

Yeah. That’s what they call the fifth metacarpals. Ask 10 biomechanists how many degrees of freedom the human hand has, and you’ll get 20 answers. Definitely. I’ve researched this. You’ll hear some people say that there’s 27 degrees of freedom in the hand. No. Now, someone who says that is probably a gamer who has a hand avatar in their gaming system with their VR and are counting the first six degrees of freedom, which is the position of your wrist and orientation in space, and with the hand DOFs. It’s like: that doesn’t belong there. So if you take that away, it means you’re getting down to maybe 21 degrees of freedom.

Scott Walter · 00:34:46

Some people might say there’s 22, depending upon the metacarpal. And what that means is that we need to do something called opposition. And that is our thumb being able to oppose other fingers and also doing it over on the pinky. And when it comes to the pinky, it turns out our pinky can kind of swing over because we have what they call this metacarpal bone down here that actually has a joint down there. And it can kind of swing over a little bit, as well as the fourth can a little bit. So there’s some arguments on whether you need both in there.

Scott Walter · 00:35:14

And that’s why you start getting into this, how many degrees of freedom that it has in there. And you can argue any number you want, but generally it’s considered to be somewhere between 20 and 22, whether you want to do that. When it comes to two degrees of freedom, I’ve heard approximately 20. And you’re like: how can they say approximately? It either is or isn’t because it’s integer, right? And it’s like: no, it turns out if you take your thumb and move your thumb, you’ll notice that none of your fingers move. If you take your index finger and move it, you’ll notice no other fingers move until you get to a point that suddenly the other one starts moving down there and there’s nothing you can do to stop it.

Scott Walter · 00:35:46

So that means there’s a loss of independence between them. And that means they start having fractional degrees of freedom. And when you add them all up, it’s something like 29.75 or something or 19.75. It’s a weird number. Now, the Sharpa hand and the Wuji hand have absolutely no coupling between them. So they are truly independent. They can move any one of their joints independent of any other. Tendon-based hands, they actually do have some crosstalk, apart from the movement of here and like another tendon’s rubbing up against another tendon. If it pulls on it, it’s possible.

Scott Walter · 00:36:18

So there is not full degrees of control as you think there’s a lot, but there’s still some subtlety that you have to tease out in your control algorithms to make sure you get your finger position where you want to have. And again, I’m devolving everything here. I would say one of the most interesting hand designs I saw recently was like a 1.3 meter hand. So I’m going to make this giant hand. And he made it like the Sharpa hand in a way, and that each joint was a big actuator. And when you get up to that scale, it makes total sense because you can size it in a way that it does not overwhelm the size of your design because you can get incredible torque density when you start scaling up.

Scott Walter · 00:36:54

And so a tendon-based hand almost doesn’t make sense when you make them big. So if you want to have like this big giant sculpture thing, just build them like a robot arm and don’t worry about doing tendon drives or anything like that or cable drives or anything like that. Well, the only alternative might be like a hydraulic-based hand because hydraulics are great when you start scaling up, like in excavators. And you can kind of fit them in there that suddenly they’re not dominating the size of whatever it is you have. When you start scaling down, suddenly your hydraulic pumps and everything become massive compared to what it is you’re trying to move.

Scott Walter · 00:37:29

So you can kind of look at it that way. And what ends up happening is as you scale down, unfortunately, the torque density just doesn’t go down the way you want, which is why you have to start considering something like a tendon-based hand versus direct drive at human scale. And it’s like right there, right at that point. If you went way smaller, really, really small to the micro level, Maybe direct drive makes sense. But kind of there’s like an area in there which you might say, yeah, remote actuation is the only way to do it.

Scott Walter · 00:38:01

And it seems like tendons make sense at that point, not so much hydraulics as you get bigger. Okay. This is more like chapters one and two as opposed to a very simple prologue of what you want to talk about.

00:38:14 — Allonic — braided hands and portfolio disclosure

Michelle Sun · 00:38:14

Yeah. Well, do you want to dive into Allonic? I think this is a really good segue to when you mentioned—I need to look up the 1.3 meter hand. And whenever we design something, I think designing for the extreme always teaches us something unexpected, right? So that’s why I want to dive into three different extreme designs that we don’t see a lot and they are very surprising. And maybe you can share what it teaches us in terms of a new perspective and what’s possible. So Allonic—you told me about it, actually. Tell me more.

Scott Walter · 00:38:48

First thing we want to do is that we’re talking about a lot of different hands and a lot of different hand companies. And Allonic is like the only one we’ll be talking about, which is actually in the RoboStrategy portfolio. And so I just want to put that out there, full disclosure. This is a RoboStrategy company and it’s a very unique design. As you can see, it’s a tendon-based design, but it’s a little bit different than your standard tendon-based designs. And that is that almost everyone is trying to figure out how to build a joint mechanically.

Scott Walter · 00:39:19

And they make a little tiny door hinge and they’ll have a tendon going on there that’s pulling it to open and close. And a lot of people have been studying different ways of being able to do that including Tesla was trying to come up with a novel way that they seem to have abandoned and gone back to the same thing that everyone’s just going to make a simple pin joint with a tendon there. Now, if you look at the human body, there’s only really one joint, which we can see a true mechanical analog to, and that is like our shoulder joint and our hip joints, which is a ball and socket joint.

Scott Walter · 00:39:51

Look at it and go, oh, that’s just, just like saying. But we’re all familiar with a door hinge. You look at the human body and I think even biology in general, I don’t think you will actually find biology having coming up with what you consider a pure door hinge. But there’s an analog in the way they work and that they have what’s called a rolling contact joint. They move around like that and the constraints are such that you can emulate it very easily with a door hinge or what we would call a pin joint or a cylindrical joint, just like that.

Scott Walter · 00:40:18

But the reality is it functions very different. So, again, when we go to degrees of freedom, I could argue that your IP joint has six degrees of freedom. You’d be huh, what? Well, the dominant one is in this direction, but it does have a little bit of wiggle in a bunch of different directions. And our body tries really hard to make sure there’s only one dominant direction. And a door hinge does a very good job of like eliminating all the other kind of motions, though it still has a little bit of slop depending on how tight it is there.

Scott Walter · 00:40:50

But the thing is, the door hinges mechanically lock it in to make sure that doesn’t move around. Now, if I have two things that are just doing rolling contact like that how do I keep it together? How do I keep it from falling apart? And the way the human body does it is we have these things called ligaments. So basically just imagine you get a bunch of rubber bands that are holding it on in there that will allow it to stretch. But you get the rubber bands in such a way that if you try doing that it just pops you right back in there.

Scott Walter · 00:41:17

But it’s a line that will allow you to do that very easily. And then you use tendons to do the actuation. So, what’s the difference between a ligament and a tendon? A tendon is basically muscle to bone attachment. And a ligament is basically bone to bone. And so you have that in there and they’re allowed to stretch a little bit. And what’s kind of great is that they act as spring sometimes. It’ll help pop everything in there. So, the human body has figured out how to do it in a miraculous way. And there’s also a little bit of a saddle joint in there to help keep a little bit of movement this way.

Scott Walter · 00:41:48

So, what Allonic has been trying to do is solve a couple of problems. We talked about tendon wear, right? The tendons wear out. And they fray all the time. Well, wait a minute. It’s like: us humans, we’re tendons. And we don’t seem to have that problem. And it’s like: we actually do because we have this thing called sleep. And when we sleep at night, our body is like regenerating and fixing everything that’s damaged. So, if you build like a tendon-based hand, you’re going to have a maintenance schedule. Every now and then, you’re going to have to replace the tendons.

Scott Walter · 00:42:20

Now, we don’t replace the tendons. We are able to miraculously repair the tendons. And if you really want to do the comparison, we want robots to go 24-7. And if we say, ah, the robots can’t go 24-7. The tendons are everywhere. We can go 24-7. Like, no, we don’t. It’s like: we work an eight-hour shift, and we’re exhausted. And if we continue to work eight-hour shifts nonstop, our tendons will fray and break. They need a break. And sometimes, they need a weekend or more. And if you’re not careful, you need surgery. And sometimes, you’ll have these cases that they never solve, that will end up having something with their carpal tunnel syndrome or something like that.

Scott Walter · 00:43:03

Or if they’ve ever had tendonitis. No, it takes a while. So we also have to be very careful how we play with the use of our tendons to keep everything nominal. So think about, a surgeon is known for his hands, right? That if they’re smart as he is, it’s all about his hands. And surgeons are not bricklayers. They will not do that. Because if they go out and start becoming bricklayers, they will lose their surgical skills in a way because they are stressing their mechanical components in a way that is just not going to work long term.

Scott Walter · 00:43:37

Okay, so what’s kind of the idea here with Allonic? Well, rather than repairing your hand, what do you imagine if you just dispose it and replace it again? Because a lot of times, it’s a question of cost. So when we are doing any sort of work, we put on consumables, gloves, to protect our hands. And then at some point that we’re out and we dispose of them, we put them on there because it’s cheaper to do that than expect our skin to repair itself all the time. And so, a lot of times the cost in a hand is where the actuation is.

Scott Walter · 00:44:07

It may not necessarily be out there. So the big problem with most hands is, you want to have them be dexterous, you want to have them be robust, and you want to have them be cheap. And it’s really hard to have all three. So the idea of Allonic is to go ahead and actually make them cheap by using traditional braiding technology. Now, we take our textiles for granted, but these machines are amazing. What they can do when it comes to any sort of textile, whatever they have come up with, that can just weave things so quickly in any sort of design.

Scott Walter · 00:44:41

We’ve seen embroidery and stuff like that. These machines can just do, right out there. And they recognize the same thing. It’s like: by using braiding technology that’s usually used to make ropes, they can also make fingers. And that part of it is that they do have a stiff member in there, which is a simple 3D-printed bone. So you feed the bone into the braiding machine, put these things around there, and you can determine from the pattern, and you can put in different fibers. So you can have, thicker or stronger or different kinds of fibers woven in there.

Scott Walter · 00:45:13

In some cases, you’ll see they do have, a different color fiber in there just to kind of prove you can do that. And that means you can start weaving all the complexity that would normally be very hard to assemble mechanically. And so it’s like: oh, certain kind of tendons, we can go ahead and do that. Oh, we need a couple ligaments here. Oh, I’ll put them in. What material do you want? So it doesn’t have to all be Dyneema. It could be cotton threads. It could be nylon threads. Whatever you want, which will give you the strength you want, give you the capability, the compliance you want in some places.

Scott Walter · 00:45:44

And the other thing, the difference between this hand and all the other hands I’ve had is that it feels kind of warm and comfortable. It’s soft. Everyone else’s hands are metallic or they’re plastic. And in many cases, their compliance they build in by putting a layer over it. So you can put a glove on it that makes your metal hand feel softer, but it still feels a little bit metallic. And so this, when it comes to actually coming into contact with people, will feel more human-like than many of those designs. So that’s the other thing that’s, super attractive about it and that they could just knock these things out really quickly.

Scott Walter · 00:46:22

And as a result of being able to do that you don’t care about the fact that your hand wore out. You just replace it because it now becomes a consumable that’s on the order of magnitude of maybe the normal kind of consumables you have.

00:46:37 — Allonic — cost, maintenance, and control

Michelle Sun · 00:46:37

So that’s definitely a very different approach. And you mentioned that there’s this trade-off triangle of robustness and dexterity. What’s the third? The cost, right? So this one, what is the BOM cost for this? And what do they sell it for?

Scott Walter · 00:46:57

So the BOM cost will probably be in the order of $100. So when they finally get the scale, this is going to be remarkably cheap. Who knows? It could be less than that. You’d love to really get it down the cost of gloves. I know some gloves are pennies, right? It takes gloves people throw on. There are other gloves in the order of a couple of dollars. So it’ll still be a little bit more. But in the grand scheme of things, if you can get a couple of weeks of wear out of this, it doesn’t matter.

Scott Walter · 00:47:23

And it depends on your application. It could be that this is going to last very, very long. But if you put it in a very dirty environment, or something where you’re very, very aggressive with it, it could be something that wears out very quickly. But at the same time, if you’re in an environment that has a lot of grit and everything, chances are the protective gear that the people are wearing is also very expensive. And it’s just so different and unique of everything else. There’s still challenges: how do you control this thing? Because it’s tendon-based.

Scott Walter · 00:47:54

But everyone is sort of getting over and around how to figure to do it. Because with enough ML, I think you can solve these problems. Everyone’s trying to simulate how their tendons are going to work. And figuring out everything to get a heuristic kind of control of it. And my feeling is, I think if you just start collecting enough data, enough ML data, your neural net will just know what to do. Just like we do. We know exactly how to manipulate our hands without even consciously thinking about it. And we’re not simulating or trying to model this with mathematics.

Scott Walter · 00:48:34

So sometimes engineers overthink it. And this is what I’ve always seen as the power of ML. If we just get enough episodes and just train it and everything, that will be the solution rather than saying, I have to have the super high-fidelity model to be able to control it. And Allonic, they’ve come out of stealth. They’re still in the startup phase. They have a lot of samples. We’re kind of bullish on them for that particular reason. But again, full disclosure, I might be a little bit biased here. Take everything with a grain of salt.

00:49:03 — Different designs for different applications

Michelle Sun · 00:49:03

Yeah, thank you. So this is a really great first look at a very different design. And one of the things that also stood out to me is that we may need different designs for different use cases, right? So for Allonic, it’s disposable, almost like a glove, high-use environment. And it may be for scenarios that are not really precise. Precision is not super needed. It’s like: oh, picking up things and things like pushing the cart around. And so maybe that tendon-based sim-to-real gap is not as much of a deployment blockage.

Scott Walter · 00:49:46

And just to point out, I’ve been looking at so many different ones, and everyone asks, well, which side of the fence are you on? And I’ve jokingly called you’ve got the tendonistas on one side and the drive cells on the other. Where am I? And I’m like: I think the TAM is so big and the application space is so big, they’re all going to win in particular niches. They will kind of self-select to the applications that they’re best suited for, and they will thrive in that very well. Because we have yet to figure out the hand that rules all hands.

Scott Walter · 00:50:17

We haven’t collected all the infinity stones yet when it comes to being able to come up with that. So you’re absolutely right, Michelle, that you’re going to see that tendon-based hands just make sense in certain applications and direct drive and other stuff like that. And the two sides will constantly argue over who has the best. But in the end, the customers will decide and they’ll find the applications and they’ll go in there and they’ll be very well. So that’s why I’m kind of bullish on that whole sector in a way is that they all will be able to find traction where it’s needed.

Scott Walter · 00:50:48

And then everyone’s like: oh, that’s a cop-out. It’s like: come on, choose one or the other. And it’s like: no, the reality is I think you don’t have to choose one. You only have to choose one if you decide to build your own hand. And then it turns out there’s a bunch of companies over in China that have like four or five different versions of hands that are all of the above because they also recognize the same thing. And then there are the others that are just like doubling down this is the only way forward and we want to put all our effort in.

00:51:15 — Daxo — up to 120 tendons

Scott Walter · 00:51:15

So the Daxo is really cool. Everyone argues over tendons and how many you need. And Tom Zhang is like: the more the merrier. In this case, I think he’s got up to 120 tendons. And then he thinks the more tendons makes it actually better. And he talks about it being more like infinite degrees of freedom, actually tendons and motors. So he actually has that many motors. They’re very small motors and he’s able to actually fit them in a space that’s about the size of a little bit bigger, kind of Popeye, but he’s able to get the form factor down there.

Scott Walter · 00:51:58

And he’s got so many tendons that he’s got a completely different control algorithm. Now, this was his first generation hand. And as you can see, it’s extremely flexible. I call it almost like sausage hands that there’s no clear kind of IP joint or it has way more IP joints than we’re used to. And that means as far as gestures and everything can do, it can do way more than a human hand can. And that was his first version. His second version, he actually has something that looks more like a human hand as far as IP joints going in there.

Scott Walter · 00:52:32

And Tom knows a lot about human anatomy and stuff like that. So he’s taken biomechanics. He’s also had like a startup that was working on a type of mechanical human hand. So he understands everything very well in the traditional approaches and was trying this other approach. And now he has sort of the normal number of joints. But again, he’s completely overactuated as far as the number of tendons. And what could be the potential advantage? Why would you even want to do that? Well, the human body actually has more tendons than everyone else is doing. So we think of that we have like a flexure and an extensure.

Scott Walter · 00:53:09

So Tesla did not have an extensor at first. They use a spring return. Now they’re putting one in there and they have two abductors in there. They were at first three. Now it looks like they’re going to be four and many others are four tendons per finger. Well, we actually have more than that in human anatomy. We also have these intrinsics internally that help us with the subtle control and movement. But we have more than one flexure anyways for grip strength and everything else. So looking at that well, wait a minute, maybe the more the merrier.

Scott Walter · 00:53:37

Because the more tendons you have, the more grip strength you can get when you need it. Because rather than having one actuator trying to pull on that finger, I can have multiple pulling on that. The other thing is that you have way more kind of weird combinations. So what a lot of people have been trying to do with the abduction is basically keep their cake and eat it too. It’s like they have two tendons because you need the two tendons for abduction. One pulling this way and one pulling that way. And then they’re running kind of a flexure through here.

Scott Walter · 00:54:10

But rather than having that flexure move this joint, they route it in such a way that it only does the PIP and the DIP. And what ends up happening is that if you pull both abductors at the same time, you don’t get ab or adduction. It actually forces you to get flexion. And that’s like a real clever thing. It’s like: oh, look, we can get this two for one here. And if I only pull on one, I get this and that. The problem with that is you actually lose a little bit of control of where it is, which is why you actually need an active extensor back there to give you the antagonism you need to put it in there.

Scott Walter · 00:54:40

And the way I want you to understand what’s happening here is that a lot of people are familiar with horseback riding. And you have reins on a horse and those leather reins are just like tendons. They pull, they work in one direction, but when you relax it, you can’t push it, right? You can kind of put it out there, but the horse is not going to feel that it’s only going to feel tension this way or that way. And we all know to make the horse turn its head this way or that way, we just have to pull on the rein this way or that way and how to return it.

Scott Walter · 00:55:09

And I want you to think of that just as abduction back and forth, back and forth. You can do that.

Michelle Sun · 00:55:14

So it’s like a pulley—there’s two.

Scott Walter · 00:55:19

It’s a pulley mechanism. Exactly. But only one of them actually controls and the other you just release to make sure you’re not putting the tension back. Now, if I pull on both reins at the same time, the horse’s head can’t go left and right. There’s only one thing you can do and that is it has to come back up. So that would be like getting in flexion. So you look at that and say, look, I get all the control I want. I can put the horse’s head wherever I want. If the horse is cooperating, what if the horse is like upset and it’s just like moving its head around?

Scott Walter · 00:55:49

Suddenly you realize you don’t have degrees of control to keep it where you want to because it’s fighting you. It’s bucking against everything you’re doing. And so what’s that got to do with fingers? Well, what happens when fingers come in contact with something? As soon as it comes in contact, I have to resist that. Otherwise, my finger is going to go. I want my finger to stay there. Well, as soon as I do that and start fighting that thing, suddenly I am inducing another movement somewhere that’s very hard for me to control because with just horse’s reins, I don’t have enough degrees of control.

Scott Walter · 00:56:17

You would have to actually add something else in there to be able to do that to give you that kind of stability. And these are the problems with minimizing the number you have. Now, if you go overboard and you put tons of them on there, suddenly not only can you get extra strength you want, you can get the stiffness when you need to have, and also different degrees of control because he’s got enough going to each group that he can move each pinky or each IP joint independent of another, which is something that’s very difficult to do when you have a minimum number in there.

Scott Walter · 00:56:49

It’s like you get that coupling that’s very hard to do. Now, most people can’t do it. Common’s an unusual person that I keep on saying, yeah, but people, can’t, you get this coupling and suddenly he’ll show me he can do it. I’m like: I don’t know how he does it, but he’s got like control of his fingers in a way that most people don’t.

00:57:07 — Daxo — grip strength, sensing, and proprioception

Michelle Sun · 00:57:07

But what kind of use case do you think this unlocks? Like this kind of hand can do that other hands that are less actuated cannot do?

Scott Walter · 00:57:16

Okay. Potentially, it could give you a lot more grip strength, which is the problem that everyone talks about is ah, we’re kind of lacking the grip strength, kind of dexterous control, resistance to force. But the other thing that’s rather interesting in here, and this is where it comes to ML, is that a lot of fingers, we’re trying to figure out how to get tactile sensing in there. Because this is something we haven’t really talked about is we’ve got all these movements. What about the tactile? It’s like: oh, that’s another episode. But it would be nice to kind of know where the touch is.

Scott Walter · 00:57:47

And there’s ways of doing haptics or sensing. One of them is that we actually have like the pressure sensors that we can feel something, which is amazing. But we also have this other kind of sense of where our fingers are, which is known as like proprioception. And we can also feel pressure and resistance that isn’t necessarily being felt through our fingertips, but we can kind of feel it in our forearms. So a lot of times, you’ll notice if you push on something like that you might notice that something in your forearm is tensioning up a little bit.

Scott Walter · 00:58:18

You can feel the touch there. So if you were to actually numb your fingers so that you can’t feel anything, you will notice you can still tell when you come into contact with something because that your forearm goes into tension. So that means there’s data out there about the contact you’re coming within. And it’s very, low quality signal, but it’s still signal and information. The other thing that all the tendon-based hands have, and this is a problem with the Tesla is because if you kind of go back to the Tesla image, you’ll notice there was like a lot of blinking lights and other things that were in there.

Scott Walter · 00:58:50

And you might want to know, what’s that all about? You see those green lights in there? Well, the problem again with a tendon-based hand is as it moves and comes into contact with something, I now have no idea what the angle of my joints are. With the direct drive guys, they know exactly. They can basically count revolutions on their fingers and they know where they are. They don’t actually need to have a decoder. I think they may have some built in there for some kind of redundancy. But the tendon-based ones, it’s like you’ve got no choice.

Scott Walter · 00:59:18

If you want to know where it is, you need some sort of feedback to be able to do that, which means ideally you’d like to do it just by measuring the tendon length or how much your actuator down here rotated. But the reality is, that’s kind of a guess on where it is, but the reality is it might be somewhere else and you want to know how to compensate it. Well, one way is I guess I just kind of look at it and estimate the angle. But in many cases, they need to actually have sensors there.

Scott Walter · 00:59:44

And a lot of those LEDs you’re seeing in there is because they have encoders built in there. And that’s basically the signal that lets it’s there and operating. I’m assuming the full-blown one won’t have that. So suddenly you’re having to build additional electronics. And like that means not only are you routing tendons, you’re routing wires through there and all these other things that kind of know what’s going on. It’s like: oh, so now I’ve got additional stuff that has to go there. So what Tom’s doing here is that he is able to tell that when you come in contact with something, that all the tendons, it’s basically kind of this idea of a tensegrity, that all of them suddenly feel attention and are reacting to that structure.

Scott Walter · 01:00:27

And the reaction to that structure, he’s feeling down here in the forearm because he has 120 motors down there. And those motors are set up that they have extremely good torque transparency. And he’s able to say, oh, this tendon, that tendon, they’re all feeling a certain force on here. And he runs this experiment enough times. Now, if you ask me to sit down and write the algorithm for that I would have absolutely no idea how to model. It’s a very difficult thing to model. Tom’s like: oh, we just run enough experiments and we know. And so he’s got it to the point that he can come up and he can touch his finger somewhere.

Scott Walter · 01:01:01

And he knows exactly where the point of contact is based on everything that’s down there because he has so much overactuated everything. So his philosophy there is that there are so many other things we can gain from doing it. Strength, cool proprioception that’s better than anyone else.

01:01:17 — Tacta — measuring how workers use their fingers

Michelle Sun · 01:01:17

Very cool. And I love that Daxo is taking the approach of maximalism, right? Let’s get as many actuators in there, as many tendons in there. And then the third example that we have is actually the opposite approach: let’s think about whether we even need five fingers. In fact, a lot of people that I’ve talked to talked about how many DOFs are really needed to handle like 90% of the tasks that humans actually do. It’s actually way less than 22 DOFs. So it seems like Tacta is an example that we can talk about also where they have five fingers, but then their launch actually showed.

Scott Walter · 01:02:03

They went along with that assumption. And again, there’s this huge academic debate over how many fingers you need and stuff like that. A lot of people like to weigh in on X about this all the time and say, well, I can do these things with three fingers and why do I need four and everything else? And again, there have been some academic studies showing that and most of the pushback they see from people is you can do a lot with three fingers and you kind of start getting out to your parade or frontiers that there’s still others that you need to have the other, have it.

Scott Walter · 01:02:36

So a lot of times it is very academic, but what Tacta did is they said let’s settle this debate and actually start measuring data with five finger gloves and they measured it from real workers in doing real tasks. So not some lab where they’re pretending to do stuff, but they actually suited up real people with a lot of tasks and they were surprised themselves to find out that the last two digits seem to be going along for the ride most of the time. And that meant that they believe that they can take over everything with just a three-fingered hand, because the others aren’t needed and they won’t have to worry too much about the body gap.

Scott Walter · 01:03:14

Now, some people say, ah, you really can’t do that. And part of the reason why we have five fingers is fatigue is that in order to generate the grip strength and everything, a lot of times we need those five finger hands. So many cases, the five fingers come in there more for a reason out of fatigue, not necessarily dexterity. And they realize this is our fingers don’t get tired and we can apply a little bit more force. So that’s why they’re able to say that they can get away without the index finger and without the pinky for that reason.

01:03:44 — Tacta — fluidic actuation and industrial arms

Scott Walter · 01:03:44

Now, Tacta is a lot of things. So the first thing is, I think they were kind of sitting there looking at the arguments and the mudslinging and everything going on between the two sides of people who are pro tendons and people who pro direct drive. And they’re like: maybe there’s a third way, a different way of being able to do the actuation. There’s kind of a hybrid between the two and they use something called fluidic tendons. And it’s a little bit like hydraulics and hydraulics have been used before. So, for instance, Clone is using hydraulics.

Scott Walter · 01:04:14

Sanctuary has done that. And the big question is: what’s the difference between hydraulics and fluidic? Well, usually with hydraulics you have one pressure source that creates a pressure and you have these micro valves that are able to open and close very quickly to be able to sort of let the hydraulic fluid in or out. And you can almost think of those as tiny actuators, but it’s more or less like open and close and figuring out how to do that really rapidly. So that’s been the trick that both Clone and Sanctuary, they have patents for things like that.

Scott Walter · 01:04:44

The other is that rather than having one single reservoir with these things that open up, why don’t you just have a syringe mechanism? So we all know how a syringe works, being able to bring up a fluid or to push it out. And so you just have a fluid and you have that actuated and you just push on it as necessary. And it also gives you a lot of control.

Michelle Sun · 01:05:02

Are the fluids currently in the palm area or where do you see it right now in this diagram? Well, in the forum.

Scott Walter · 01:05:10

Okay. So it is backed out somewhere else. So there’s a much bigger base. It’s not even in the forum, it’s in a base, but they’re going to be scaling it down. So right now it’s kind of bulky and it’s really meant to go onto a cobot. So this is the other thing. A lot of people talk about the humanoid distraction when you’re building a hand. A lot of people are trying to build a hand to go on the humanoid. And when you do that you have to do all these things. It constrains you so much on the size and the power and the weight and everything else.

Scott Walter · 01:05:36

But there’s all these robotic applications out there that just need a hand. They don’t have to be in a humanoid form. So we’re like cobots, industrial arms and stuff like that. There’s a lot of applications. If they had something like a human hand would be able to do a lot of it. And they’re not the only ones that have noticed that. Mimic has noticed that. Kyber has noticed that you can go down the list of some others. Like, wait a minute, let’s not worry about this distraction, the humanoid distraction and think about what would happen if we put it on a real robotic arm, which is already certified.

Scott Walter · 01:06:07

They have great precision. They already have a programmatic interface. We can source them really easy. Let’s do that. So that’s how they’re looking at that. And they’re not worried about how this is going to get onto an Optimus. They’re not trying to sell it to Optimus or to Tesla or to Figure or anyone else. They’re like: no, this is going to be going to other people. And they’re trying to solve that at hand problem. And the way they’ve done it is basically a kind of a cross between trying to be a bit direct drive and also be tendon-based.

Scott Walter · 01:06:37

And that instead of using tendons, they’re using a fluid to replace the tendons. And they’re using a tube, which you could argue is kind of like a Bowden tube in a way. So a lot of the routing is again done with these tube structures that will be able to go around but you can then have a wrist without the problem. Because again, if you’re having a tube going around there, you don’t have to worry about the length changing. And the fact that they have a fluid in there, the fluid doesn’t really care about length. It just cares about pressure.

Scott Walter · 01:07:03

And you can sense the pressure really quickly and react to that. So a lot of very interesting things about this hand design are using a different way of doing actuation.

01:07:12 — Tacta — gloves, data collection, and tactile sensing

Scott Walter · 01:07:12

But as you’ll see, Tacta is like kind of many companies in a way because it’s not just the hand and a few others are beginning to realize you need what literally is like hand in glove, right? You need to capture data. And there’s always this embodiment gap that everyone’s trying to figure out. Well, now how do I collect the data? Do I go egoistic, this and that and everything else. And a lot of times that’s not good enough. Do I go with an UMI glove? How do I do the mapping or retargeting from what my hands like?

Scott Walter · 01:07:42

Wouldn’t it be great if I could just slip this hand onto my hand and collect data with it? That’s the ultimate embodiment and or the ultimate kind of UMI. And they have a glove, which is very similar to the hand as far as the mechanical movement. And as far as the touch sensing, because that’s the one mode everyone says like with egocentric data, you can capture the movements, but you can’t capture the actual tactile feel.

Michelle Sun · 01:08:10

Is that the glove that you’re referring to on the screen?

Scott Walter · 01:08:13

Yep. So they have a glove, which is able to pick it up. And when they’re doing that they are also capturing what the tip sensors are going to feel. Now, the sensors they have all over, they came up with a way of being able to mass produce very high precision and high density taxels. So these are basically tactile cells. Think of them like the way we think of pixels or voxels, except for touch. And they have them in all the key locations, including the palm. So they know what it’s like to grab something. And I’m pretty much of that feeling that the data collection we have to do has to have a lot of modes and that a lot of the data, if it doesn’t have some sort of touch or force feedback, the data will be a very low quality.

Scott Walter · 01:09:03

It’s still usable. But what it means is you probably need a much larger data sets to be able to overcome that deficit. But if you can actually capture that mode right away, you’re way ahead of the game, way ahead. So they already were the philosophy. It’s like: we’re not just going to build a hand. We have to build the data collection thing to go with it. And we’re not just building a hand. We’re also building really good touch sensors that are going to be in there. So that’s some other part of the IP. That’s very interesting approach.

Scott Walter · 01:09:30

So it’s like the full stack of the hand in a way. And they’re also having to build the model that goes on top of it. So very serious company based in Palo Alto, actually just down the road from where the Tesla headquarters is. And when I visited them last week, I was really quite impressed with the operation that they have there. The number of engineers that they have, the equipment that they have in there is just wow, these guys, they kind of know what they’re doing.

01:10:00 — Tactile response rates and sensitivity

Michelle Sun · 01:10:00

Yeah. And one of the things that you talked about, the touch sensor part, I read that they have 400 hertz, right? So I think a lot of the numbers that I’m seeing is more than like the 50, 100 hertz. What does the 400 hertz unlock versus a lower number?

Scott Walter · 01:10:18

Yeah, that’s very important. We always hear these numbers that humans are able to react in like a 10th of a second. And I know that from like track and field, because they always said that it takes a 10th of a second for the runner to react to the sound of the gun. And that’s how they actually tell a false start, believe it or not, is that when the gun goes off, if the pressure that’s on there happens before a 10th of a second, it’s considered a false start, which is rather amazing, which means you cannot anticipate.

Scott Walter · 01:10:47

So it’s kind of based on human reflexes. And we always think about that. And a lot of times the update rates, I think our vision systems update also somewhere around 30 hertz, maybe it might be technically 20, but I think it’s considered somewhere around there. So everyone’s always built to that. And the assumption is that’s human reaction speed. But it turns out our touch sensing is incredibly reactive and it’s closer to a thousand hertz. So we’re getting a lot of information really quick. And it seems a lot of that is not so much going all the way back up to our brain, but it’s more like reflex action that we have.

Scott Walter · 01:11:23

So we have a lot of kind of edge compute or neural compute built into our hands and our reflexes. So you notice if you touch something sharp, you react way before you’re aware of it or something hot, and that’s because we are able to sense very quickly and they’re trying to get it down to that human scale level. The other thing they’ve done is that they can actually sense pressures that are a quarter of what humans can sense. So it’s even way more sensitive as well as way up on the high end. And as with a lot of instrumentation, either an instrument is really good at something fine or it’s really good at something big and it’s hard to get the two to be able to do both and they’re able to change.

Scott Walter · 01:12:05

That’s impressive. So they’re trying to get down to those levels. And that was new information to me, that there’s actually part of the human nervous system that is able to react faster than 100 hertz in the order of a kilohertz is amazing.

01:12:19 — Industrial use cases and customer discovery

Michelle Sun · 01:12:19

And one of the things that I’m trying to reconcile is that it’s really great that Tacta is tackling the hand space in a more deployment-centric view. Right. So instead of thinking to sell to humanoids, actually, why not? We just build a habit for the COBOS systems and go to the industrial setting. So in terms of this very fast tactile response and also sensing different pressure in a very precise way, what is the use case in an industrial setting? I get that at a home setting, if I touch something really hot, then I want to react.

Michelle Sun · 01:13:01

But in an industrial setting, what are some use cases that will be very useful?

Scott Walter · 01:13:11

They’re all sorts. The one use case I see coming up again and again and again is data centers and cabling and tables and data centers. And like they’ve probably had to plug in like an internet cable in the back of a laptop or something like that. Data centers are like that but on steroids and slightly different kinds of connectors. And a lot of times knowing when a connector is getting it in requires sometimes a lot of finesse because sometimes you can’t see it. So you have to be able to kind of react, tell from the field.

Scott Walter · 01:13:39

And then when it clicks in, know that you’ve got the right kind of click. So we are very good at doing that. And with the scale up of data centers, everyone is looking at trying to automate that. So that’s like kind of one example. You also have just like a lot of assembly, like consumer electronics and stuff like that. The finesse that kind of goes on. And then it turns out, see, this is the other thing that Tacta is able to do with their fluidic gloves. They can actually make smaller size hands. Most of the other hands, they’re bigger than most human hands.

Scott Walter · 01:14:11

So when I put my hand up there, it’s like the size of my hand. A lot of people complain about the size of my hand and higher. And in China, in a lot of electronics assemblies, most of the people who are working there are women. And a lot of times they’re there because they’re petite and their hands are much smaller at that scale that actually allows them to do it. I would have difficult time because my hands are both thumbs, kind of what it comes down to. And so they can actually scale it down to the size that will allow them to do that.

Scott Walter · 01:14:40

A lot of these very precise motions that are just very hard to automate in general. And again, what they’re seeing is that a lot of these assembly plants are people sitting at their desk. And all they’re doing is moving their arms. You don’t need the legs and you barely need a torso, but you need this fine dexterity control that’s out there. And the equipment is already there. It’s like: oh, they grab this tool. They go ahead and do that. They grab something else and be able to go. So they’re finding that. And this is what I encourage all entrepreneurs out there.

Scott Walter · 01:15:11

And I’m seeing a lot of them that are more like that. Actually get out to the deployment phase early. I would say even pre-deployment. Going out, meet your customers, find out what their needs are, and then design everything around that. And you will definitely be more successful way sooner than everyone else. If you’re sitting in a lab, everything looks great. And then you go to deploy. And then suddenly you’re discovering all sorts of things you did not know were going to be problems. And now you have to go back and redo it. So find your customers.

Scott Walter · 01:15:42

I’ve met a couple entrepreneurs, and it’s just great. They literally just go on a field trip. Someone will just pack up the car and drive around and just knock on doors of all these different like small companies to find out what they’re doing, what their manufacturing challenges are. And some cases find, the manager that’s there or the owner being really excited about, yeah, we’ve been wanting to automate this thing over here. These are the challenges, the problems, and the guys learn about it. And then they go back and they build something and then they get traction.

Scott Walter · 01:16:09

That’s the way it’s going to be. The TAM is so big. Don’t worry about building hands for humanoids. I am very pro-humanoid. It’s going to be there. But I don’t take the side that it’s going to be special purpose versus general purpose. There’s a huge spectrum in between there and that you’re going to need both because there’s just so many applications out there. And if you try to get kind of dogmatic that only one side is going to work and you think you’re going to dominate the whole thing, no, no, no, stay full.

Scott Walter · 01:16:44

That is the biggest pie we’ve ever seen, ever, trying to automate labor. And the tiniest sliver of that pie will keep you satisfied for your entire life. You won’t be able to eat that piece of pie. Let’s just put it that way.

01:17:04 — General-purpose bodies, specialized work

Michelle Sun · 01:17:04

Exactly. And if we think about humans, right, we’re general purpose, but we also have different professions. So a surgeon is different from an assembly worker. The skills, they’re all precision-type skills, but it’s a very different type of expertise as well. And the part that you mentioned really going out to the field and the deployment site, like it was really so true where when I visited factories in Shenzhen, I’m like: wow, these people, they literally sit there for 12 hours a day and they like have a basket on top of them. And then they move things and assemble and then move it to another basket.

Michelle Sun · 01:17:39

And so they really only get up to use the bathroom or get a sip of water. And so I think these are the insights that makers like Tacta would think about: hey, if we really want to be useful, maybe put the energy to make a really sensitive tactile sensing system. And then a very dexterous hand, and maybe think about how the movement is not as important as sometimes we want to think of.

Scott Walter · 01:18:05

Yeah. And the other thing to kind of point out is, this argument about, one humanoid can do everything. It’s like kind of the humanoid form factor can, but we’re going to see different versions of it for good reasons. And that is, we know, they’re different sized people and some of them excel in certain areas and others just because they’re a little bit taller, they’re a little bit smaller. Look at the Olympic Games. Gymnasts tend to be on the short side. Basketball players tend to be on the tall side. And a lot of it’s just physics.

Scott Walter · 01:18:41

If you’re tall, it’s really hard to do cartwheels. I can tell you that I cannot do a cartwheel. Okay. I am jealous of all these people I’ve seen doing cartwheels. So you’ll see these different embodiments make sense. And a lot of times it’s just that again, a surgeon kind of follows a certain path and that it may be that because they decided to not be a bricklayer, they have the hands to be a surgeon. But if they had made that choice earlier in life, they wouldn’t be able to be a surgeon because they’ve just kind of lost that finesse that you need.

01:19:17 — Why the next five years might be different

Michelle Sun · 01:19:17

Yeah. I think one last question that I have is, people have been saying that robotics is five years away for the past almost 30 years. And so what’s on the table now that wasn’t available 10 years ago that you think will truly unlock the next five years being that inflection point? Is that the edge compute side or is it more like the sheer amount of data and maybe it’s a bit of both?

Scott Walter · 01:19:54

Well, I think the most obvious thing is just machine learning and AI has changed everything dramatically. So you can do a lot more. First of all, the fact that we can just use that to control robot armors versus the old way, which was pretty much just that it would do the same thing. It didn’t have any intelligence to adapt to any changes in the environment. So that’s just opened up so many possibilities. Now, the reason why I say that’s important: there’s all these other things that are important. The other is just the interest in the field.

Scott Walter · 01:20:29

Ten years ago, no one was really interested in robotics. Now, that’s all everyone wants to do. And the more minds, especially brilliant minds, you start throwing out a problem like this, the more likely you’re going to solve it. The other has been the result of a lot of improvements in mechanics. And a lot of it comes because people are trying to solve the problem now. They’re trying to solve problems that could have maybe been solved before, but they weren’t interested in. So, making smaller actuators and stuff like that. We know how to make motors for the longest time, but no one really thought of making motors down to that scale that you could put in humanoid.

Scott Walter · 01:20:59

So, you saw Elon complain about that five years ago. That you could not find any actuators on the market for any price to do what you want. Now you can. It’s incredible. You go to Amazon, you can order and you get them a couple of days later. So there’s that. There’s the fact that locomotion is a solved problem, whereas like 10 years ago, it was still very challenging. So all these things are happening: this convergence of technology, new material sciences, new people trying to solve all those things. So what’s happened is that there’s just the excitement.

Scott Walter · 01:21:35

Maybe it’s just this big paradigm shift and the realization that humanoids are possible. Physical intelligence is possible. And this is something I’ve said before people. It’s like I’m a broken record on this. But one of my favorite quotes is from Arthur C. Clarke, where he says the space elevator will be built 50 years after everyone stops laughing. And I’ve paraphrased that to humanoids. Humanoids will be deployed like five years after everyone stops laughing. And I think everyone has kind of stopped laughing now. So that five-year we’re talking about—we’re seeing them already. Come on. They can do some manipulation tasks.

Scott Walter · 01:22:17

They’re just not quite reliable yet. They are going into some deployments. They are doing some of the low hanging fruit. And they’re the worst they’re ever going to be. They’re just going to get better and better and better. And five years from now, that’s a lot of time in humanoid space. Really, when I go back thinking about when we saw the Unitree bot doing the RL program walking, I think that’s ages ago. And I realized that was only 18 months ago. It was like January 2025. And so it’s like: oh, I thought that was, again, in my head, it’s like five.

Scott Walter · 01:22:51

But so what we’re seeing, and again, if we look at the humanoid games and also the mini marathon, the Beijing mini marathon, the contrast in 12 months is just unbelievable. And you’re going to see like that same step change next year in both of those. I’m really, really pretty sure you’re going to see it. So five years, yeah, we’ll be there. We’ll be there. And humanoids will be quite performative at that point.

• What task does the robot perform?

• Which part of the system is holding it back?

• What evidence proves it works repeatedly at the required economics?

I built the task-first Physical AI stack as my framework for answering these questions.

It maps 11 functional layers, from the part that touches the work, through actuation, power, intelligence. It can be used to locate a bottleneck in a system, to understand where a company competes, and to follow where the value accrues.

What is the physical AI stack

The physical AI stack is the set of hardware, software, data and operating systems that turns a goal into a reliable action in the physical world.

Most technology stacks are drawn from the enabling technology outward: chips, compute, models, applications. For robots, I find it more intuitive to begin with the task. I call this the task-first stack.

Start with a job such as placing a product in a bin. A gripper touches the product, an actuator moves the gripper, the power system supplies the required energy, the control software regulates the motion, planning finds a feasible path, perception locates the object and bin, a policy selects the action, and data trains the model.

There’s also compute, simulation and evaluation, which we’ll cover in this piece.

These 11 layers map the working system behind a robot task. Beneath them is the industrial foundation: energy, critical materials, manufacturing capacity and physical infrastructure.

The task-first stack changes the starting question. Instead of asking how advanced a model has to be, ask what has to go right for the task to succeed. The answer then points us to where the bottleneck is.

One task through all 11 layers

Recently Stanford student and OpenAI Robotics intern Thijs Simonian gave GPT-6 Astra an SO-101 robot arm, a paintbrush and a camera, then asked it to paint the Golden Gate Bridge. Astra planned one minute of actions at a time, reviewed the result and adjusted across attempts with feedback.

This is a good example of how the stack works and identifies the bottleneck. We will also go over 3 additional examples at the end of the piece.

For this Golden Gate Bridge painting task, the bottleneck is the control and actuation interface between Astra and the robot. The model (GPT-6 Astra) could recognize what needed improvement turn after turn, but the robot’s control interface, actuation and limited feedback limited how precisely the arm could execute the model’s judgment.

Don’t worry if the table doesn’t make sense to you right now. We’ll walk down the stack one layer at a time. After that, refer back to the table and it will make more sense.

1. What is a robot end effector? Task meets the world

A robot’s end effector is the part that touches the work. Common forms are a parallel-jaw gripper, vacuum cup, magnetic tool and multi-finger hand. Its geometry, sensing and materials determine which objects and processes the robot can physically handle.

The market spans task-specific industrial tooling and generalized dexterous hands. SCHUNK, Robotiq and OnRobot sell grippers and end-of-arm tooling. Shadow Robot, Wuji Hand and Sharpa develop multi-finger, or dexterous hands. I wrote about the designs and economics in Dexterous Hand Primer:

An end effector needs to manipulate objects at the required speed, force, tolerance, uptime and price. A simple gripper is sufficient for pick-and-place tasks. A dexterous hand widens the task set, but also adds joints, weight and cost.

Tactile and force-torque sensing sit close to this layer, or you can think of sensing as layer 1b. These sensors tell the robot whether contact is secure, excessive or beginning to slip.

What matters: what touches the work, and what tasks does that choice open up or rule out?

2. What are robot actuators? Turn command into force

Actuators convert electrical commands into movement. The actuator shapes the force, speed, precision, efficiency and thermal performance a robot can achieve.

Humanoid joints have to reconcile high torque, low weight, compactness, low backlash, reliability, cost and manufacturability.

I saw this layer up close in a precision gear factory in Songgang, Shenzhen. 26 wire-cutting machines ran around the clock. Staff inspected parts to tight tolerances. The joint’s eventual force, precision and reliability began in processes like these.

An actuator that is heavy, expensive or thermally constrained can limit the system even as the software improves.

What matters: can every joint deliver the torque, precision and thermal margin the task requires, at a weight and cost the robot can carry?

3. What powers a robot? Store and deliver energy

The power system provides energy to the actuators, compute, sensors and auxiliary systems. For mobile robots, this usually begins with a battery pack and extends through the battery-management system, power electronics, wiring, protection and charging interface.

Capacity determines how long the robot can work between charges. Pack weight competes directly with payload, while charging time and cycle life shape fleet utilization and replacement cost.

The power system can constrain runtime through charging frequency, battery swaps and thermal limits.

What matters: can the power system sustain the task’s peak load? What is the runtime of the robot?

4. What is robot control? Move toward the target

Control is the software layer that converts a desired motion into the continuous stream of motor commands that produces it. The control loop recalculates the required position, velocity, torque or motor current hundreds to thousands of times a second on a real-time processor. It has to absorb a late measurement, a small collision, a heavy object or a battery-voltage change while keeping the robot stable and within its limits.

Two common approaches are (1) PID, proportional-integral-derivative, the classical feedback loop: it measures the gap between where a joint is and where it should be, then corrects in proportion to that gap; and (2) model predictive control, or MPC, which simulates the effect of potential commands and picks one that satisfies constraints such as torque limits, balance, collision clearance or energy use.

Much of this layer is available through open-source libraries and vendor tools, including ros2_control, Drake and NVIDIA Isaac. Robot companies still have to integrate and tune the loops for their mechanics, tasks and safety limits. Companies such as Applied Intuition and Intrinsic also sell parts of the broader robotics software and tooling stack.

What matters: when something unexpected happens, what keeps the robot safe and on task?

5. What is robot motion planning? Choose a feasible path

Motion planning is software that selects a feasible sequence of movements, from the robot’s current state to a goal. For a manipulator, that may include choosing a grasp, or placing the object without colliding with the shelf. For a mobile robot, it may include route planning and task sequencing.

Classical methods include graph search, inverse kinematics, sampling-based planners such as rapidly exploring random trees, and trajectory optimization. MoveIt is an open-source manipulation framework. MoveIt Pro packages a commercially supported workflow. NVIDIA’s cuMotion library accelerates trajectory optimization on GPUs.

Planning and control are adjacent but serve different purposes. Planning decides what movement to attempt. Control makes the robot follow it in real time. This guide traces the dependencies from the object, so control appears first. During the task, information generally flows from perception to planning to control.

What matters: can the robot find a safe, efficient path for the current scene, and recover when that path fails?

6. What is robot perception? Estimate what is happening now

Robot perception turns sensor readings into an estimate of the robot and its environment. Cameras, depth sensors, LiDAR, force-torque sensors, tactile sensors, encoders and IMUs all contribute different information. A camera may identify a package. A wrist force sensor may reveal that it is snagged. Tactile sensing may show that it has begun to slip.

Perception answers a present-tense question: where are the object, robot and relevant contacts now? A model can use that information, together with instruction and goal, to decide what to do next.

Some perception functions are available as libraries, including object detection, pose estimation, visual odometry, SLAM and sensor fusion. NVIDIA Isaac ROS is a packaged example. For deployment, the robot still depends on sensor calibration, latency and edge cases of the target environment.

Control, planning and perception usually run on the robot. Perception and planning commonly run on the main onboard computer, while the fastest control loops may run closer to the joints.

What matters: what does the robot need to observe, measure and infer before it can act reliably?

7. What are VLMs, VLAs, world models in robotics?

Models interpret observations, reason about a goal, predict outcomes or generate actions.

• A vision-language model, or VLM, processes visual and language inputs. In robotics, it can interpret a scene and an instruction and produce a high level plan.

• A vision-language-action model, or VLA, combines vision and language inputs with an action output. Physical Intelligence’s π models, Google DeepMind’s Gemini Robotics and NVIDIA’s GR00T are examples.

• A world model learns how a scene may evolve. Given a current state and a possible action, it predicts the future state. NVIDIA Cosmos and World Labs’ Marble are different approaches in this broad category.

• A world action model, or WAM, is a world model that also outputs actions. Dyna Robotics’ DYNA-2 is one example. NVIDIA’s Cosmos 3 also includes policy variants that connect video prediction with robot actions.

Perception estimates the present, a world model predicts possible futures and a world action model ties those futures to executable choices.

The model is absorbing the stack. These models are increasingly taking on work that previously sat in the perception and planning layers. For example, a VLA may generate a high-level action chunk directly from camera input and an instruction, bypassing several hand-built perception and planning interfaces. It may also generate lower-level actions, but fast motor control still sits underneath it. A world action model can predict how candidate actions change the scene, which can support or replace part of a planner’s work.

What matters: which decision does the model make, and which decisions remain with conventional software and control?

8. What data trains physical AI systems?

There are 5 main types of robot-learning data:

• Robot demonstrations (or teleoperation data): A human controls a robot and records observations and actions together.

• Human demonstrations (or egocentric data): Humans record themselves completing tasks.

• Simulation: A virtual environment generates trajectories, labels and failures under controlled conditions.

• Internet and video data: YouTube and other online videos provide demonstrations at much larger scale than robot fleets.

• Deployment data: success, failure and recovery data from robots doing the actual work

I wrote about robot data here.

Volume is only part of the equation. The data also need to be relevant to the task, embodiment, sensor and environment. Failure data and variety of data are also key.

A whole universe of companies is forming around this layer. XDOF builds robotics data infrastructure and datasets. Mecka describes itself as a data, evaluation and deployment layer for physical AI, with a focus on egocentric human activity.

What matters: who can produce data that is relevant to this robot’s work, and how does that data improve the next release?

9. Where does physical AI compute run?

Robotics uses at least three kinds of compute:

(1) Cloud or datacenter compute trains models, processes fleets of data and runs large simulation workloads away from the robot.

(2) Onboard accelerated compute handles latency-sensitive perception, model inference and sometimes planning. NVIDIA Jetson and Jetson Thor are examples of this tier. Qualcomm, AMD and specialized edge-compute vendors also make processors for this tier. These computers need to fit the robot’s power, heat, weight and cost budgets.

(3) Real-time embedded compute executes fast control and safety functions on motor drives, microcontrollers or real-time processors. Texas Instruments, NXP, Infineon, STMicroelectronics and motor-drive vendors make processors and control components for this tier.

What matters: which computation must happen on the robot, and can the hardware deliver it inside the latency, power and thermal budget?

10. What is robotics simulation used for?

Simulation creates a controlled environment for design, training, integration and stress testing before a policy reaches the real world.

Simulation can generate variation that is expensive or unsafe to capture on hardware. It can also test an integrated stack before a physical robot is available at scale.

The sim-to-real gap is the mismatch between the simulated environment and the physical one. Contact physics, material wear, lighting, sensor noise are common sources of mismatch. A policy can perform well in simulation, and still fail in a real deployment.

MuJoCo is a physics engine widely used in research. Gazebo supplies open-source physics, rendering and sensor simulation. NVIDIA Isaac Sim connects simulation with NVIDIA’s broader robot-development workflow.

What matters: what did the team validate in simulation, and which real-world conditions remain outside that environment?

11. How should physical AI systems be evaluated?

Evaluation measures the gap between reality and expectation. For example, whether the system works across changing conditions, recovers from errors and makes economic sense. A useful scorecard includes task success, intervention rate, recovery behavior, cycle time, latency, energy consumption, safety and cost per successful task.

Evaluation also closes the learning loop. Failure helps inform the model training, data collection, hardware or workflow adjustments.

What matters: how do we tell that this system works in the target environment and business case?

Where does value accrue in the physical AI stack?

Companies create the most value when they solve the bottleneck holding back an important task.

Better models reduce the burden on hand-built perception and planning. Better sensing, end effectors, actuators and power systems make more behaviors possible or keep them running longer. More deployment produces more interaction data. That data then improves the model.

The cycle is better robot → more deployment → more data → better model → better robot.

The flywheel turns only as fast as its slowest layer. For one company that slowest layer is data collection. For another it is the hand, the actuator, the power system, edge compute, safety validation or the cost of running the robot at all.

The most valuable part of the stack can change as technology improves. Today, the main limitation may be the model and data pipeline. As models improve, the bottleneck may move to the actuator, sensor, or deployment network.

How can investors and operators use the physical AI stack?

Start with a specific task and ask 3 questions:

a. Where does the system learn? Trace who generates interaction data, who owns it and how it reaches the next version.

b. What limits the task today? Locate the constraint in the model, sensing, mechanics, power, reliability or unit economics.

c. What evidence would establish deployment? Ask for the operating conditions, task success, recovery behavior, cycle time, energy use and the economics of repeated operation.

The task-first stack breaks down a robot demo into a chain of parts that can be examined individually. Trace the task from physical contact back through the stack. It is then easier to see what the system depends on, where a company may have an advantage and where the robot is most likely to fail.

For paid readers, below I apply the stack to three tasks and surface three different bottlenecks: warehouse bin picking, inserting a connector and loading a dishwasher.

Applying the stack: three tasks, three bottlenecks

Read more

Welcome back to the 4-part series “the Business of Actuators”. This is part 2. If you’re new here, this series aims to cover the business of actuators. We go deep enough technically in order to understand the underlying factors driving the economics of actuators, and hence physical AI. Part 1 covered what is inside a humanoid joint: mechanical parts, the main designs, joint anatomy and the low-ratio vs. high-ratio design trade. Start there if you missed it.

This piece looks at how much an actuator costs. We work from an audited filing, current vendor listings and an independent teardown. One buyer’s audited price for a single driven axis is $202, and an independent teardown estimates a comparable joint near $167. [1]

And thank you for the feedback on the first post. If you are building actuators, robotics components or industrial automation systems, I would love to hear from you.

Sections

1. How Much of a Humanoid’s BOM Is Actuation?

2. The Right Unit of Comparison: Cost Per Actuated Axis

3. What Does One Actuated Axis Cost?

4. Why Retail Pricing Overstates What a Manufacturer Pays

5. What the $202 Benchmark Actually Tells Us

6. Data Provenance and Methodology

1. How Much of a Humanoid’s BOM Is Actuation?

One benefit of the current wave of Chinese robotics IPO filings is that we are finally getting more granular supply-chain data.

LeJu Robot, the company behind the Kuavo humanoid, filed for an IPO on the Shenzhen Stock Exchange’s ChiNext board under its fourth listing standard for high-growth technology companies.

Its prospectus reports direct-material cost of $19,587 per Kuavo unit, representing 78.9% of total build cost. [1] Most percentages below use that $19,587 direct-material denominator; a few use total unit cost or another filer’s own hardware-cost base, each labeled where it appears.

Start with the joint modules. Kuavo carries between 26 and 38 modules. LeJu’s audited procurement table gives a blended module price of RMB1,115, or approximately $156. That puts joint modules alone at 20.7% to 30.3% of direct-material cost. [1]

Now add the driver electronics. LeJu buys the driver boards on a separate line. Driver-board spend is 29.5% of joint-module spend in the same audited table, which puts $46 of drive electronics on a $156 module. The combined module and driver basket therefore reaches 26.8% to 39.2% of direct-material cost. [1]

Then add the dexterous hands. LeJu’s procurement data prices the pair at $3,370. Including the hands brings the full electromechanical drivetrain to 44.0% to 56.4% of direct-material cost. [1]

For a full electromechanical drivetrain including hands, the share is 44.0% to 56.4% of direct-material cost and 34.7% to 44.5% of total unit cost. [1]

Quick glossary:

• Direct materials are the parts and inputs that go into the robot.

• Total unit cost is direct materials plus assembly labor, manufacturing overhead and other costs allocated to the finished unit. For reference, 78.9% of Kuavo’s total unit cost of $24,825 is direct materials.

• Electromechanical drivetrain is the full motion path: joint modules, electronics and hands

• Joint module is just the mechanical assembly, motor and reducer.

This is why published actuator shares are all over the place. These calculations are using different inputs: the numerator, the denominator, and the robot. [1]

• Numerator: the actuator can mean at least three things: (a) the mechanical joint module, (b) the joint module plus driver electronics, (c) the full electromechanical drivetrain including the dexterous hands.

• Denominator: which dollar figure you divide by. Direct materials, total unit cost, hardware cost and ex-factory price are four different bases. The same $156 module is 0.8% of Kuavo’s $19,587 direct-material cost and 0.63% of its $24,825 total unit cost, before the numerator basket even changes.

• Robot: which machine produced the numbers. A full-size biped, a wheeled AMR and a quadruped carry different actuated-axis counts and different architectures, so the share moves with the machine even when the basket and the denominator are held fixed. Youibot’s 50.2% below states its numerator and denominator clearly, and it is still the wrong robot for a humanoid benchmark.

Our own earlier estimates of 40% to 50%, and separately 50% to 65%, labeled these boundaries too loosely. I am retiring them in favor of the basket approach above.

Drivetrain BOM share has trended up over time. Youibot, which makes wheeled and inspection robots and filed for an HKEX listing under Chapter 18C, reports drivetrain rising from 33.7% to 43.8% and then 50.2% of its hardware cost over three years. [2]

Youibot (a Chinese industrial robot company)’s fleet is still mostly AMRs and inspection robots. It launched two humanoids in 2025, a wheeled model in March and a bipedal model in August. Humanoid revenue stays inside total company revenue in the filing and reads as immaterial. As a reference, Youibot shows that the drivetrain can become a larger share of the hardware basket as robot capability increases.

For a full-size biped the share is higher. An independent teardown of the Unitree G1, which shipped more than 5,500 units in 2025, puts its joints alone at about 66% of that robot’s bill of materials. [8] That covers joints only, on a different robot than the Kuavo ladder above.

2. A Consistent Unit of Comparison: Cost Per Actuated Axis

Before comparing prices, we need a consistent unit. Robotics terminology makes this more confusing than it should be.

• A joint module is the physical mechanical assembly around a joint: typically the motor, reducer, encoder, bearings and housing. The driver is the power and control electronics.

• An actuated axis is one independently driven direction of motion: one commandable output with its associated motor and transmission chain.

• A physical joint can contain more than one axis. A wrist that both bends and rotates may contain two independently controlled axes in a single joint assembly.

• Actuator is the physical hardware component.

• Degrees of freedom, or DoF, describe the robot’s kinematic freedom rather than how many motors it contains. It refers to the total number of independent, movable joints that a robot can shift position or rotate.

Actuator count, actuated-axis count and DoF are often close to one another. But they are not universally 1:1.

Below are three examples:

Tendon-driven dexterous hand. To actively control 1 single rotational axis in both directions, 2 actuators are needed to pull in opposite directions, like a pulley. This is similar to bicep and tricep muscles for humans. (actuators = 2, actuated axis = 1, DOF = 1)

Swinging pendulum. A swinging pendulum has unactuated (passive) degrees of freedom (DOF), but zero actuated axes. (actuated axes = actuator count = 0, DOF = 1).

Soft robotics hands. Made of silicone, the material can bend in every millimeter of its structure. So, its DOF is theoretically infinite, while driven by a single cable motor (actuated axes = actuator count = 1, DOF = ∞).

For consistency, we use cost per actuated axis as the comparison unit in this piece. It gets us closer to the cost of adding one independently controlled direction of motion to a robot.

3. What Does One Actuated Axis Cost?

LeJu’s filings shed light on actuators volume pricing in China today.

Its audited prospectus reports Rmb40.3mn ($5.63mn), spent on joint modules in 2025 at a blended unit price of Rmb1,115, or about $156 per module. [1]

Dividing total module expenditure by that disclosed unit price implies roughly 36,000 modules purchased. LeJu buys driver boards on a separate line. Driver-board spend is 29.5% of joint-module spend, which puts $46 of drive electronics on each module. [1]

That brings the combined mechanical module plus driver electronics to:

$156 + $46 = $202 per actuated axis.

This is a useful high-volume benchmark. Zooming in to the data and recent statements, we found three insights about the pricing:

1) Price halved as LeJu’s purchasing scaled 100X. Over the two years LeJu’s annual module purchases went from roughly 370 to roughly 36,000, and its per-axis cost fell from $433 to $202, with nearly all of the decline on the driver-electronics line. [1]

2) $202 is a related-party price. LeJu owns 5.73% of the supplier making these modules, its finance chief sits on that supplier’s board, and roughly 83% of its comparable-product procurement comes from that related party, Wuxi Quanzhibo. [1] The filing says pricing uses a cost-plus arrangement and sits slightly below comparable products. [1]

3) Industry-grade joint modules on average cost $210 in 2025, to fall 10% in 2026 and mass-market pricing at $140. Quanzhibo, LeJu’s supplier, shared some details on pricing in July. Its CMO shared the 2025 industrial-grade joint-module average near Rmb1,500 ($210), expects module costs to fall roughly 10% in 2026, and the whole-robot mass-market pricing at Rmb1,000 ($140). [9] LeJu’s Rmb,115 blended price sits below that 2025 industrial average, which is consistent with the filing’s “slightly below comparable products”.

The filing gives no torque, mass, gear ratio or gear family for the $156 module, so it represents an average across joint sizes. One independent build-cost estimate exists for a joint of the same type. China Post Securities’ March 2026 teardown of the Unitree G1 built its joints up from components: Rmb1,000 ($140) for a small joint and Rmb1,500 ($210) for a large one, about $167 blended, driver board included. [8]

A merchant module price of $202 and an in-house build-cost estimate of $167 are referring to the same joint module (motor + reducer + driver + encoder).

4. Why Retail Pricing Overstates What a Manufacturer Pays

Now compare that $170-200 volume figure with what a robotics startup pays one unit today.

Current online-retailer listings, quoted in US dollars, from September 2026 include:

• DAMIAO DM-J8009-2EC: 9:1 low-ratio planetary architecture, 20 N·m rated / 40 N·m peak torque, 896 g, dual encoders, onboard driver — $385. [6]

• MyActuator RMD-X8-25: $450 without the USB-to-UART accessory and $480 with it. The selected version is listed “Without Driver,” so it is not directly comparable with DAMIAO without separately pricing the motor driver. [7][11]

• ZeroErr eRob 80T: 50:1 to 120:1 strain-wave architecture, 21 to 31 N·m rated and 44 to 143 N·m peak, 1.94 kg — starting at $1,440. The listing does not clearly specify driver status.

Pendulum Robotics’ public comparison table lists 130+ named rotary actuators with price, torque, dimensions, ratio, gearbox type and encoder configuration. The table includes a $325 DaMiao DM-JH11, a 52 mm, 51:1 harmonic actuator, and a $257 DaMiao DM-J6248P-2EC, a 76 mm, 48:1 harmonic actuator. These are public single-unit retail listings for finished actuators. The $202 benchmark above is an OEM volume-procurement cost. [7]

These are not apples-to-apples products. Each quoted price bundles architecture, size, electronics integration and purchase volume

The ZeroErr strain-wave module sells at 3.6X the price of the DAMIAO planetary module, and delivers less peak torque per kilogram at every ratio, 0.8 times DAMIAO’s. We covered the performance trade-offs behind those architectures in Part 1.

LeJu’s $202 combined module-and-driver cost is about 52% of DAMIAO’s $385 retail price for a broadly comparable low-ratio module. [1][6]

Single-unit retail pricing is a poor proxy for what a manufacturer pays at volume. LeJu bought roughly 36,000 modules in 2025 across about 1,000 robots. Unitree, which shipped more than 5,500 humanoids that year, designs and integrates its joint modules in house. [8] Part of the gap between $202 and a $385 retail listing is purchasing scale, and part is the related-party pricing already described.

5. What the $202 Benchmark Actually Tells Us

We draw two conclusions from the numbers.

Read more

Actuators can account for up to 70% of a humanoid robot’s bill of materials. A typical humanoid packs dozens of them. A design decision at one joint cascades down the robot.

The key question running through this piece is: where does a robot pay for torque?

More gearing shifts the engineering burden toward precision mechanics. Less gearing shifts it toward the motor, electronics and thermal system. Repeated across dozens of joints, that single design decision ends up deciding what the robot costs to build and which suppliers capture the value.

Actuators are complex machines. Over the past few weeks I have spent a lot of time learning how they actually work. I’m thankful to the authors of the many resources I have relied on along the way. They are cited in the Appendix.

By the end of this piece, you’ll know what is inside a humanoid joint, why it is built the way it is, and how that design choice redistributes value across the supply chain, from a small set of precision gear-cutters to a global market for rare-earth magnets.

This is Part 1. Subscribe to get the future posts in your inbox as they go live.

Sections

1. What’s Inside a Humanoid Joint?

2. Why Gearing Is the Key Architectural Decision

3. Five Actuator Metrics That Matter

4. What Gearbox Choice Tells You About the Supply Chain

5. Why QDD Changes Actuator Economics

6. Where Different Actuator Architectures Get Used

7. Where the Robot Pays for Performance

8. Acknowledgements

9. Appendix A: What’s Inside an Actuator

10. Appendix B: How to Read an Actuator Spec Sheet

11. Appendix C: How the Three Gearbox Architectures Work

1. What’s Inside a Humanoid Joint?

Before we look inside the actuator, I want to clarify three terms that repeatedly come up: joint module, motor driver and actuator. The industry uses them interchangeably, but they are actually different things.

A joint module is the mechanical assembly: motor, reducer, encoder, bearings, housing, and on some designs a brake. Two filings classify the components differently: one groups the encoder inside the joint module, the other treats it as a separate control component alongside the driver board.

A motor driver is the power and control board. It turns battery DC into three-phase current for the motor and closes the control loop.

An actuator is the broad category word that can cover both. Two vendors quoting an actuator can therefore be quoting two different bills of materials, one with the driver board inside the housing and one selling that board as a separate line.

For full details of each component, see Appendix A of this post.

2. Why Gearing Is the Key Architectural Decision

Electric motors and humanoid joints naturally operate in very different regimes. Motors are most compact and efficient when spinning fast. Humanoid joints need to move relatively slowly while producing turning force, or torque.

A reducer, or the gearbox, bridges the two: it reduces the motor’s speed for greater torque at the joint.

A higher reduction ratio gives the joint more mechanical leverage, but that comes with costs. Here are three of the penalties, and how each translates into economic trade-offs elsewhere in the robot:

1. Energy loss means bigger batteries and shorter runtime. Gears lose some energy to friction. A single planetary stage returns 95%-98% of the motor’s work, a strain-wave stage drops to 85 to 90%. A stacked two-stage planetary train (we explain more what that is in Appendix C) degrades further: squaring that range gives roughly 90% to 96%, an arithmetic estimate rather than a measured figure. Across dozens of joints, those losses mean more heat, shorter runtime, and potentially larger batteries or additional cooling.

2. Heavier joints increase BOM costs. Reducers contain precision steel components. And steel is very heavy. A heavier knee doesn’t merely cost 300g more. The hip actuator then needs more torque to move that knee, which means bigger motor, more copper/magnets, stronger bearings, more current and potentially a larger battery. Mass at each joint compounds through the robot.

3. High reduction lowers force transparency and dynamic response. High gearing mechanically isolates the joint from the motor. That makes the joint harder to move from the outside. It also makes the motor less able to feel external forces. One reason is reflected inertia: the motor’s rotational inertia is amplified through the gearbox by the square of the gear ratio.

This leads to one of the central design decisions inside a humanoid is, where do you pay for torque?

And it has implications of where the cost sits in the actuator.

At a simplified level, gearbox designs can sit on either end of the spectrum:

a) Smaller, faster motor + high-ratio reducer
b) Larger, high-torque motor + lower-ratio reducer

At the high-reduction end (a), more of the engineering burden sits in the reducer: precision gears, tight tolerances, specialized manufacturing and, depending on the architecture, multiple reduction stages.

At the low-reduction end (b), more of it shifts into the motor and electrical system: magnets, copper, higher current, electronics handling motor current and thermal management.

So, this design choice determines which components matter most, which suppliers are difficult to replace, and where value accrues inside the actuator supply chain.

3. Five Actuator Metrics That Matter

These are the five metrics to look at when evaluating an actuator claim, and what each tells us about the robot behind it.

1. Continuous vs peak torque: Can the robot sustain the demo workload?

Peak torque tells you what a robot can do briefly. Continuous torque tells you what it can do repeatedly in deployment.

2. Torque density: How much actuator mass is required to produce useful force?

Higher torque density means more torque for the same actuator mass. That cascades into lighter limbs, smaller upstream actuators and potentially lower overall BOM.

3. Backdrivability: How naturally can the robot absorb and sense external forces?

This matters for walking, manipulation and human interaction.

4. Backlash: How much mechanical uncertainty exists when direction reverses?

This matters for precision manipulation.

5. Shock tolerance: Can the hardware survive repeated real-world impacts?

This matters for reliability, maintenance and useful life.

The full list of actuator metrics is at Appendix B.

4. What Gearbox Choice Tells You About the Supply Chain

The three major gearbox architectures solve the same basic problem, but their manufacturing requirements create very different supplier economics.

Harmonic drives provide compact high reduction and near-zero backlash, but require specialized precision manufacturing. This includes grinding a flexpline that has to flex millions of cycles without cracking.

That manufacturing base has historically concentrated in Japan, and a few players such as Harmonic Drive Systems and Nabtesco. Chinese entrants, including Laifual and Leaderdrive, are now taking share, and the shift looks more like a maturing competitive base than a pure price war: Laifual’s most recent interim results showed shipments and margin both rising, not falling. We’ll cover more on Part 3 of the series.

Planetary drives prioritize efficiency, shock tolerance and manufacturing accessibility. The gears supply is not scarce; it can draw from a much broader conventional gear-manufacturing ecosystem. The challenge is to hold tight tolerances at high volumes. Low-ratio planetary is also the architecture behind QDD, which trades gearbox difficulty for another supply chain problem, covered in the next section.

Cycloidal drives serve high-load applications. The hard part to source is the eccentric disc’s precision profile, machined by a narrower set of shops than planetary gear-cutting, though a broader one than harmonic. They are less common in humanoids today, though at least one Chinese humanoid maker names a cycloidal joint as a core technology, so “heavier, so avoided” does not hold across every design.

The design, pros and cons of each gearbox architecture are covered in detail in the Appendix C.

5. Why QDD Changes Actuator Economics

If you follow humanoids coverage, you probably have come across the word QDD quite a bit. QDD, or Quasi-direct drive in full, is an actuator design philosophy, not a gearbox type. It refers to a system where a high torque motor is paired with a very low reduction gearbox (3:1 - 10:1). Quasi means almost: the motor is almost directly driving the joint, with just enough gearing to increase torque while preserving much of its responsiveness to external force.

QDD gearboxes are predominantly planetary due to their low cost and high mechanical efficiency (95%+), though low-ratio cycloidal variants are emerging.

The MIT Mini Cheetah actuator, developed by Ben Katz, became an influential reference design for modern QDD modules. Mini Cheetah used a single 6:1 planetary stage delivering 6.9 N·m continuous torque and 17 N·m peak. By contrast, traditional industrial robot wrists run gear ratios between 50:1 and 100:1 to statically lock loads. Even humanoid knees, which require higher torque to hold deep crouches, use low-ratio multi-stage planetary setups (eg, 22.5:1), staying well below industrial standards to preserve shock tolerance.

QDD changes where the robot pays for torque

Traditional robot architectures often pair smaller, fast-spinning motors with high-ratio reducers to generate torque at the joint. More of the engineering burden therefore sits in the reducer.

QDD flips this. More of the torque comes from the motor, which means, more magnets, copper, and more demanding electronics and thermal management, in exchange for a simpler, lower-ratio reducer.

By shifting the workload to a powerful, high-diameter motor, QDD delivers three mechanical advantages:

1. High backdrivability. At a low ratio you push the output and the motor turns. The robot feels external force through the joint itself, enabling soft compliance. In contrast, a high-ratio gearbox behaves closer to a one-way valve.

2. Low reflected inertia. Low reflected inertia allows a humanoid to react more quickly to disturbances, such as a stumble.

3. Shock tolerance. A walking robot hits the ground several times a second. Lower-ratio designs generally tolerate impact better.

That mechanical shift has economic implications.

• High ratio keeps the burden in the gearbox: precision-ground teeth, tight tolerances, and a small set of manufacturers who can hold them. High ratio buys precision at the cost of a narrow gear-cutting supplier base.

• Low ratio pushes the burden into the motor and its electronics instead: a wider stator, more magnet volume to generate torque directly, more copper in the stator windings, and a driver board built to handle it. Low ratio buys shock tolerance at the cost of rare-earth magnet and higher-current electronics.

Even though electronics is more complex, QDD can lower the total actuator BOM of a robot. Below are some reasons:

• Supply Chain Accessibility. Low-ratio planetary gearboxes are standard industrial components. They can be made by widely available gear-hobbing equipment. Harmonic reducers need a much more specialized supplier base. That being said, a lower-ratio design needs more concentrated magnets to make up for the less sophisticated gearing. China accounts for 94% of global sintered neodymium-iron-boron magnet production, up from about 50% two decades ago. [12]. We’ll cover in Part 3 of this series

• Cost & Efficiency Gains. According to SemiAnalysis’s June 2026 G1 teardown, low-ratio planetary gearboxes achieve 95% to 98% efficiency (vs. 85% to 90% for strain-wave) and up to 80% cheaper.

• Driver Localization. Public financial filings from Leju, a humanoid maker in China, show driver expenditure per joint module fell 83% in two years.

6. Where Different Actuator Architectures Get Used

Humanoids do not use a single “one-size-fits-all” actuator. Different joints optimize for different combinations of force, precision, impact tolerance, mass and packaging.

Legs and hips prioritize shock absorption, rapid acceleration, and heavy force production. To protect from ground impact, these joints lean heavily toward QDD planetary setups or high-force linear actuators.

Wrists and arms prioritize spatial precision, high torque-to-weight ratios, and compact form factors. They often employ harmonic reducers to maximize holding torque in tight spaces. Alternatively, QDD motors are placed centrally, then power is routed through tendons to keep the limbs light.

Current humanoid designs show how these trade-offs play out in practice:

• Tesla Optimus Gen 2/3 uses harmonic reducers across rotary body joints (shoulder, torso, wrist roll) for compact form factor and precision, while it uses linear actuators for high-force joints like knees and elbows.

• 1X’s NEO hand uses a tendon-driven Quasi-Direct Drive (QDD) architecture at approximately 5:1 to 15:1 gear ratios. Motors are housed in the forearms/torso to keep distal limb inertia minimal and limbs safe around humans.

• Unitree highlights its core architecture in investor briefings as a “high-bandwidth force-controlled quasi-direct-drive planetary rotary joint”. It combines a high-torque motor, low-ratio planetary gearhead, and rapid torque-control loop. (See Unitree’s 3-hour Investor Q&A for details).

7. Where the Robot Pays for Performance

This piece we covered the key question to ask when looking at actuators for humanoids or any robots: where the robot pays for performance. High reduction puts more of the burden into precision mechanical components. Low reduction shifts it toward motors, magnets, electronics and thermal management.

That architectural choice flows through the entire robot: how it moves, what it costs to build, which components become critical, and ultimately which suppliers capture value.

On Friday, we’ll put dollars against those choices: what a humanoid actuator actually costs, using audited filings and factory quotes, and why published estimates put actuators anywhere from 21% to 70% of a humanoid’s BOM. [1][4]

Acknowledgements

Thank you to Parth Ingle (@parthingle_x), Grant (@grantg07), and Sajiv Shah (@sajiv_shah) for technical review and feedback on earlier versions of this piece. Any errors or interpretations that remain are my own.

Appendix A: What’s Inside an Actuator

Below are the key parts inside an actuator and what they do.

Stator. The stator is the fixed half of the motor: a stationary ring of copper coils. When electricity passes through these coils, it creates an electromagnetic field.

Rotor. The rotor is the turning half. Permanent magnets (sintered neodymium-iron-boron) are bonded to a steel structure. As these magnets interact with the stator’s magnetic field, they force the rotor to spin.

The motor is built primarily from laminated electrical steel, copper wire, magnets and epoxy.

Reducer. Also called the gearbox, it reduces rotational speed and turns that into greater turning force, or torque, at the output. It is typically built from hardened steel components. There are three main types of reducers used in robotics, which we’ll cover later in the Appendix. The choice of gear family can be a major driver of joint cost.

Encoder. The encoder tells the robot’s computer what angle the joint is at, how fast it is moving, and which direction it is turning. It can use a magnetic ring or optical disc read by a sensor. Some joints use dual encoders, one on the motor and one on the output.

Bearings. Bearings support the actuator’s rotating components and carry loads between the actuator and the robot’s limb. They typically consist of rolling elements, such as balls or crossed rollers, running between two hardened races.

Driver board. The driver board turns commands from the robot’s main computer (e.g., an edge compute device in the chest or head) into controlled electrical power for the motor. It typically contains power transistors, a microcontroller and connectors.

Housing and brake. The housing holds the actuator assembly together and helps manage structural loads. It also has to deal with heat generated by the motor. Some joints include a brake to hold position when required.

Appendix B: How to Read a Humanoid Actuator Spec Sheet

Let’s cover the key terms to understand when reading an actuator’s spec sheet. One way that’s easier for me to understand these mechanical terms is to link them back to the human form and think about our own joints, muscles and movements.

Gear Ratio (or Reduction Ratio) (N:1) is the mechanical trade-off between speed and torque inside the actuator. It is usually written as N:1 (e.g., 9:1, 40:1). A high-speed, low-torque electric motor turns a gearbox. The gearbox slows down the rotation at the output shaft by a factor of N, while multiplying the motor’s torque at the output by approximately N, before accounting for efficiency losses.

Gear ratios are all about mechanical leverage.

• High ratios (40:1 to 100:1) function like a low gear on a bicycle. They provide greater mechanical leverage for producing and holding torque, but reduce output speed.

• Low ratios (3:1 to 9:1) function more like a springy, dynamic athlete’s limb. They are responsive and fast but provide less mechanical leverage, so the motor needs to do more of the heavy lifting.

Peak torque vs. rated / nominal torque (N·m). Rated / nominal torque is the torque a motor can produce continuously without exceeding its thermal limits. Peak torque is the higher torque a motor can produce for a short period before heat becomes the limiting factor. A large gap between peak and nominal torque means the actuator cannot sustain its headline peak performance continuously.

Torque is raw rotational strength. A bodybuilder’s bicep generates torque to curl a 100-lb dumbbell. It answers: “How much turning force comes out of this joint?”

Torque density (N·m/kg) is turning force per kilogram. It is an important metric in robotics because mass at the end of a limb costs torque at every upstream joint. Vendors rarely list this number directly; you calculate it by dividing torque by total actuator mass. Today’s merchant joint modules typically yield 45 to 75 N·m/kg at peak. Always check continuous torque density too, as thermal limits keep sustained performance below peak figures.

Think of an elite rock climber vs. a big powerlifter. A rock climber might be able to lift less absolute weight than the powerlifter, but can generate high force relative to body weight. In the same way, torque density tells us how much rotational force an actuator produces relative to its mass.

High torque density is important for robots because a heavier actuator requires more torque just to move the limb itself. It is also a major engineering challenge: producing more torque generally requires more motor material, which adds mass.

Backlash (arcmin) is mechanical slop, often measured in arcminutes. One arcminute is one-sixtieth of a degree. Turn a gear train one way, then reverse it, and the output can briefly remain stationary while the teeth cross the gap between them and begin transmitting force in the opposite direction.

Think of backlash like looseness in a joint. Too much can cause instability and reduce accuracy in fine tasks like threading a needle.

Very low backlash is useful for precision tasks, although some gear designs require small clearances for lubrication, thermal expansion and reliable operation.

Backdrivability describes how easily an external force applied at the output can drive the motor backward through the transmission. It can be characterized by the amount of output torque required to move the joint. A highly backdrivable joint can respond more naturally to external forces.

The analogy: a backdrivable joint is like a relaxed arm. When someone pushes your wrist, your arm yields smoothly. A poorly backdrivable joint behaves more like a rigid arm that resists being moved from the outside.

Backdrivable designs are useful for human interaction, force control and dynamic movement. Less-backdrivable designs can be useful when stiffness and holding loads are more important.

Shock tolerance describes how well an actuator survives sudden impact loads, such as when a robot’s foot hits the ground during walking. Vendors may characterize it using peak or momentary torque relative to the rated figure. The merchant modules reviewed above range from about 2x to 4.6x rated torque.

Appendix C: Comparing Gearbox Architectures: Harmonic, Cycloidal, and Planetary Drives

Below are the three main gearbox architectures used in robotic joints.

Harmonic, cycloidal and planetary gears all transfer rotational power and reduce speed. The difference is in how they do it. The three architectures make different trade-offs between positioning accuracy, shock tolerance, weight, efficiency and manufacturing complexity. Below, I’ll use the same human-anatomy analogies to help visualize each architecture.

Harmonic Drives (Flexing Cartilage & Tendons)

• How it works: A harmonic drive uses a flexible, thin-walled steel cup that deforms inside a rigid outer ring to transmit motion.

• Anatomy analogy: Think of flexible cartilage or tightly tensioned tendons. The architecture relies on controlled elastic deformation rather than only rigid gears rotating against one another.

• One-liner: Premium precision at a cost.

• Strengths: Very low backlash and high reduction ratios (often 50:1+) in a compact, lightweight package. This makes harmonic drives useful for applications such as precision industrial and surgical robots.

• Drawbacks: Higher unit cost, specialized manufacturing requirements and lower tolerance for shock loads than some alternative gearbox architectures. The flexible component is also subject to fatigue over its operating life.

Cycloidal Drives (Interlocking Vertebrae)

• How it works: A lobed disc moves eccentrically inside a ring of stationary pins or rollers, distributing forces across multiple contact points.

• Anatomy analogy: Think of interlocking spinal vertebrae. Loads are distributed across multiple contact points rather than concentrated on a small number of gear teeth.

• One-liner: The rugged option for heavy loads.

• Strengths: High shock tolerance, high load capacity and low backlash. Multiple contact points allow the gearbox to distribute sudden loads across the mechanism.

• Drawbacks: Cycloidal drives tend to be heavier and bulkier and can be more complex to assemble. This makes them more suitable for high-load applications where ruggedness matters more than minimizing joint mass.

Planetary Gearboxes (Skeletal Ball-and-Socket Joints)

• How it works: A central “sun” gear drives multiple “planet” gears inside an outer ring gear.

• Anatomy analogy: Think of a shoulder or hip joint: a rigid structure that distributes force across multiple contact points.

• One-liner: The scalable, cost-effective workhorse.

• Strengths: High energy efficiency (often 95%+ for a single stage), good shock resistance, and an established manufacturing base using conventional gear-making equipment. Low-ratio planetary gearboxes are commonly used in Quasi-Direct Drive (QDD) actuators at ratios such as 3:1 to 10:1.

• Drawbacks: Planetary gearboxes have mechanical backlash. Stacking multiple stages to achieve higher gear ratios can increase backlash, friction, weight and complexity.

Share Core Matter

Appendix D: Sources Cited

[1] LeJu / Lejoin Intelligence ChiNext prospectus, audited by Tianjian, procurement tables pp. 1-1-187 to 1-1-188.

[4] SemiAnalysis, “China’s Unitree Will Dominate Global Robotics,” 2026-06-09, read via Core Matter’s own capture of the piece.

[12] International Energy Agency, “China’s share in rare earth magnet production, 2024,” Rare Earth Elements report, April 2026: https://www.iea.org/data-and-statistics/charts/china-s-share-in-rare-earth-magnet-production-2024

What we cover

• Why fingers break before motors or gears in a robotic hand

• The move from tendon-driven transmission to rigid linkages in Prensilia’s product line

• A 300,000-cycle grip force test protocol and what it does not tell you

• Underactuation, degrees of freedom, and degrees of actuation explained

• Bill of materials breakdown for a 3-motor dexterous hand

• Customer mix across research, prosthetics, and industrial buyers

• Manufacturing scale from hundreds of units a year toward volume production

“Fingers are the parts that break the most, because you have impacts with objects, you have unexpected movements from the robot.” — Francesco Clemente, Managing Director, Prensilia

“We are talking about thousands of cycles, not really millions of cycles, with the tendons.” — Francesco Clemente, Managing Director, Prensilia

“The motors account for maybe thirty five percent, more or less, of the total cost.” — Francesco Clemente, Managing Director, Prensilia

Francesco Clemente on LinkedIn
Prensilia Website
Watch on YouTube

Subscribe now

Every episode, I sit down with the founders, researchers, and operators building physical AI worldwide, tracing the full stack from components and supply chains to what actually deploys. See all published episodes here

Chapters
00:00 Francesco Clemente, Managing Director, Prensilia
02:22 Prosthetics and research customers
04:46 Broken hands and reliability
07:09 Loaded versus unloaded cycle testing
11:59 Failure modes in fingers, gears, and motors
14:19 Motor heating and cooling
16:42 Tendon maintenance and anchoring
21:26 Weight and robotic-arm payload
23:48 Degrees of freedom and actuation
26:14 Underactuation and adaptive grasping
31:01 Motor current, torque, and heat
33:29 Grasp taxonomies and motor count
35:52 Abduction and hand design
38:17 Tactile sensing
40:38 Bill of materials and actuators
45:22 Industrial use cases
50:04 Competitors and grippers
54:42 Manufacturing scale

Core Matter is an independent research practice covering the physical AI value chain

Michelle Sun: Fingers are the parts that break the most. And that connection is a weak point that is gonna break. The motors are bottlenecked, and having a lot of motors inside of a small volume, it’s easy to look at spec sheets, but you have to understand that Francesco Clemente, managing director of Prensilia. Prensilia builds robotic hands out of Pisa, Italy. It’s fun out of set and in a school of Advanced Studies in 2009. Started off in prosthetics, today their hands are on humanoid robots, factory arms, and human wrists. Prensilia’s Mia hand runs three motors, closes in 280 milliseconds, which is faster than a human hand, and has survived 300,000 cycles at full grip force in testing. It sells for $10,000 to $15,000 compared to the Chinese hands at 5k. On the spec sheets, the Mia hand might look overpriced and under threat. What they’re seeing in order volume says something different. Today we’ll talk about what’s on the ground in the dexterous hand markets.

Francesco, welcome to the show. Prensilia is 17 years old today from 2009, and you are a spin off from Sant’Anna, started way before robot hands were cool, and definitely before the humanoid boom. So what did the company look like in 2009 and who paid the bills before humanoids existed as customers?

Francesco Clemente: Yes, thank you very much for having me today. I’m very excited to be here. The company, Prensilia, was founded, as you said, by researchers of the artificial enzy area that were working at the BioRobotics Institute of the Sant’Anna School of Advanced Studies. They were already working as a researcher to on the development of robotic hands and they were using developing those tools for their own research needs. Then other researchers that they were collaborating with started asking for accessing those tools, and basically here it how it came the idea to spin off a company in order to allow other researchers also to use these devices that were developed in the lab.

So the first customers were really other researchers, so universities and research centers around the world. And that let’s say research background is still with us. So today we are still working, collaborating a lot with research centers, and this is also one of the reasons why we have the specific versions of our robotic hands for research activities.

Michelle Sun: That’s amazing. And so the Mia hand came out of a prosthetic. How is the medical version similar or different from the robotics version? Did you have to change anything for the robotics customer?

Francesco Clemente: Yes. Apparently robotic hands can be used, let’s say, for prosthetics as well as for robotics today, but these two worlds are relatively different because you have very different specifications from one market to the other one. For instance, in prosthetics, you have a very clear bottleneck on the ability of the person to control the prosthesis. So you have a certain amount of degrees of freedom of movement that the person can control in a reliable way, so it doesn’t really make a lot of sense to have a 20 degrees of freedom robotic hand that can do any gesture, because the person cannot control that complexity.

On the other side, you are also limited by the weight and by the size of the device. Of course, besides prosthetics, you have specific sizes in prosthetics that you try to match. That is like small, medium, and large, I would say. The prosthesis has to be that specific size, it cannot be any size. So today we see robotic hands that have very different sizes, because of course you attach them to a robot, so it’s not super important if the robot itself and the robotic hand match in size perfectly, let’s say. But this is very important in prosthetics.

And of course another point is weight, because the person has to actively carry the prosthesis. The comfort is very important in that case, so you cannot really have a prosthesis that weighs two kilograms, because the person will be tired after one hour, let’s say, of using the prosthesis. So you really have to go down in weight in order for the prosthesis to be comfortable.

Michelle Sun: When we first spoke, you mentioned that at ICRA in Vienna earlier this year, people have been showing up at your booth carrying broken hands from other vendors. Tell me that story again. What had failed and what were the customers looking for?

Francesco Clemente: Yes, so we were at ICRA and we were meeting a lot of people at our booth. A lot of them were researchers or engineers from companies that were looking for robotic hands. And some of them were saying, okay, we are looking for robotic hands that are robust, that can be used also outside of the lab, because we were trying some other hands from competitors, but they didn’t work. We bought them because of the low price, but then we realized basically that they were not reliable for what we were going to do.

So I think that this summarizes a little bit what the status of the market is today, because there is a lot of competition, a lot of hype around robotic hands, and everyone is really working on these devices, so the demand is going higher and higher, but people are somehow approaching these devices for the first time. So they don’t really also don’t know exactly how to use them, what are the specs that are important. I mean, it’s normal, because these devices are complex. And for us that are working in the field since more than fifteen years, it’s clear what you are going to look at. But as you said before, robotics and also prosthetics was, let’s say, a niche market for robotic hands in particular a few years ago. Now everything is exploding, so people have to be accustomed also to understand the specs and what they are getting for the money.

Michelle Sun: I wanna double click on the point you make about there are things that are not on the spec sheets, right? So there’s degrees of freedom and you can also see the comparative price pretty easily. You tested Mia to three hundred thousand cycles at full force. Walk me through that protocol, because 300,000 loaded cycles and a million unloaded cycles are very different claims, and on the spec sheet it’s not easy to distinguish. Yeah, talk me through how to read through the spec sheets more efficiently and tell the differences between different hands.

Francesco Clemente: Yes. One thing that prosthetics has taught us is that we want to develop a robotic hand or a prosthesis that then can be used outside of the lab by patients and users. And also engineers, it has to be robust. Okay, so it’s easy to look at spec sheets that basically report the number of degrees of freedom, which is the kind of measure of the dexterity of the robotic hand that you’re buying, and everyone is looking into that today because they’re looking into fine manipulation protocols and solving manipulation at a higher level.

And of course you want to have something that is similar to the human hand in terms of movements that you can perform, but you have to understand that that comes with a cost associated. So the device becomes very, very complex. You either have a very bulky forearm that hosts all of the motors, with tendons that drive the fingers and the joints, or you have very small motors inside of the hand, inside of the joints of the fingers. And this means that the performance will have to be low because of the size of the motors that you’re using. So reliability is something that you have to compromise, also with dexterity, let’s say.

So we really pointed to have, tried to achieve, high robustness with our devices, and this is the reason why we basically developed a test protocol where we would run several thousand cycles at full force in order to really understand if people could use the device at the specs that are reported in the spec sheet. Okay, because if you say, okay, my hand can perform this type of grasps and this amount of force, 50 newton grip force, you have to know how many grasps you can really do at that amount of grip force, you know? Because you can test it at no load. But of course that’s not a measurement, or let’s say a parameter, that tells you in the real world how much the device is going to last.

Michelle Sun: If we do the math for, let’s say, the cost per grasp cycle, the Mia hand is ten to fifteen thousand dollars and the hand at the ICRA show probably around five thousand. How many cycles does that other hand last? And on a per grasp cycle basis, which one is actually cheaper?

Francesco Clemente: Well, it’s very difficult to tell a number for competitors, let’s say, because you really have to get the device and test it to know for sure, let’s say. But the thing is that there are two points. One is the grip force that you can get with those devices, that is relatively lower. So in general you also have maximum performance that’s capped because of how the device is designed. So if you have more motors, usually you have a lower grip force, this means that also for single grips you have a lower performance. And on the other side, of course, if you run at full grasp, you probably get a shorter and shorter lifespan.

So I can say for the competitors, okay, they will last 100,000 cycles or whatever, but the problem is that, what they reported to me is the fact that the performance was too low to carry out a large amount of tasks. And also, in the tasks where they were using the hand, they were still breaking, even if those tasks were relatively light, let’s say. So also moving cans, empty cans, let’s say, would be a relatively problem for those devices.

Michelle Sun: When you say that a hand starts breaking, are we thinking about the motors overheating, or actually like some of the fingers fall off? Or tell me more about the failure modes when it comes to, at the end of the 300,000 cycles, or in a competitor’s hand, a hundred thousand cycles, what would be the breaking point of the dexterous hand?

Francesco Clemente: Yeah, there is a lot of stuff that can break in a robotic hand. So one of the reasons why we try to limit the amount of motors in joints, it’s also to reduce complexity in order to have less stuff that can break, let’s say. But in general, of course, fingers are the parts that break the most, because you have impacts with objects, you have unexpected movements from the robot, so they are the most exposed part, I would say. So it’s important to have a device that allows you also to replace relatively quickly the fingers or the fingertips in order to service the hand in a relatively quick way. And this is, I would say, one problem that we also have and that we solved in this way, basically with a quick-release mechanism that allows you to replace the more distal part of the finger in a quick way, from a user perspective.

On the other side, of course, when you run a lot of cycles, you start to have wear of the components. So we have gears inside of our device, and these wear out because of friction. So at some point you will have to replace some components. First, I mean, usually you don’t have a dramatic, sudden break of the component, but you start having more mechanical play in the transmission. And then at some point, of course, the component is worn out too much, so that it basically breaks.

Another point that is very important is related to the motors. So the motors are the bottleneck in the mechanical transmission for several reasons. One is the one that you were mentioning before, so one is overheating. So electric motors heat a lot, and having a lot of motors inside of a small volume basically means that you need to have some sort of cooling system.

So if you look at the shadow robotic hand, that is this anthropomorphic robotic hand that was available since twenty years, they have a large forearm where they put the motors, and then with the tendon system they run the fingers, they have fans on the forearm in order to pull down the motors. So this is a problem that was there since all the time, let’s say, so it’s not a new problem, let’s say. So if instead you are putting the motors in the fingertip or in the joints, the problem is that you don’t really have a space to put fans there. And the motors, the smaller they get, the more they heat, because the surface area of the case of the motor is smaller. So the capacity to transfer heat outside decreases a lot with their volume and surface area, let’s say. So what happens is that sometimes you have to stop using the device in order to wait for the device to cool.

This is again one other reason for us to use less motors, bigger ones, so that they have less heating issues, let’s say. Just very quickly, another point is the gearhead of the motor that also limits the torque that the motor can provide to the system. So another point that can break is really the gearhead of the motor. So if you are pushing the motor to the limit, the structural components of the motor can fail at some point.

Michelle Sun: So it seems like there are four key areas that the hands can fail. The fingers can fall off on its own. And then there’s the motors wearing off after too many cycles. And the motors overheating is another one. And the last one is about the gears.

Francesco Clemente: Right. The gearheads. So the gears that are inside of the motor, let’s say, or coupled directly to the motor. So the last one is the gearheads within the motors being worn out because of too many cycles.

Michelle Sun: It’s interesting you mentioned Shadow. That’s actually my next question. You spent a decade building tendon-driven hands, the IH2 Azzurra, 11 degrees of freedom. And then you decided to remove tendons. Shadow used tendons for 20 years and half of the industry roadmap is still using tendons. What was the reason behind this design decision? Is that the cycles, tension lost, or is it really expensive to service?

Francesco Clemente: That’s a very good question. So tendons, we think tendons are great because they provide you a lot of flexibility in terms of design flexibility. So you really can move the motors around, because you can use the tendon to transfer motion from the motor to the fingers. So you have the ability to have more freedom of where to put the motors inside of your device.

Also, one of the reasons why they were used a lot is when you have a lot of motors, so that you can really store them in a more convenient way, and transfer motion with tendons in a very simple way around. But on the other side, the main problem of the tendons is, first, the fact that they usually break. I mean, tendons themselves are not a problem. I mean, tendons are also used in airplanes for controlling some components of the wings, so they’re not a problem by themselves. But the problem is that when you put them inside of a robotic hand, you have a very small amount of space available to make a tensioning system that is robust enough to be used correctly for a lot of cycles.

And specifically, tendons need a lot of maintenance. So the problem is again not with the tendons themselves but with how you anchor the tendon to the transmission. So the anchoring points are usually weak points. You can make a knot, you can use sleeves in order to connect the tendon to the mechanical transmission, let’s say, and that connection is a weak point that is gonna break. So even if the tendon is, let’s say, rated to hold one hundred fifty newtons, after a few cycles it will hold probably a hundred newtons. So they’re not very good with a large amount of cycles.

So we are talking about thousands of cycles, not really millions of cycles, with the tendons. And this is one of the main reasons why we moved to gears and links. Of course, gears are more reliable on this scale, I would say. They have the disadvantage that you have less freedom, because you cannot have a gear of any size, of any shape, you cannot transfer motion at any angle, let’s say, so you are more constrained. So you can do less, let’s say, with these tools.

Michelle Sun: So where does the anchor points fit? So you mentioned that there are all these tendons, and then there are points that are the failure points, right? The anchor points are usually where the problems start to appear. Are they within the knuckles area in each of the fingers, or where do you usually hide the anchors?

Francesco Clemente: We have one hand that is called IH2 Azzurra that is driven by tendons. So these tendons are anchored to the mechanical transmission on two sides. On one side is in the palm, because in the palm we have the motors, so we have a mechanism that basically pulls the tendon. So on one side the tendon is connected there in the palm. And on the other side is at the fingertip. So basically the tendon runs through the joints of the finger and then anchors to the back of the fingertip, so that when you pull the tendon, the finger will close. And when you release the tendon, the finger will open because of a spring inside of the joints. Usually, let’s say by generalizing, one point is before the joint that you want to move, usually close to the motor, and the other point is beyond the joint that you want to move. So at the fingertip, or let’s say between the metacarpophalangeal joint and the proximal interphalangeal joint, if that joint is the one that you want to move.

Michelle Sun: I noticed that the Shadow hand with the forearm units included is around four point three kilograms. Versus the Mia hand, with three motors inside the palm, is around a tenth of the weight. Walk me through the math when it comes to the weight and the performance that comes with it.

Francesco Clemente: Yes, of course. So weight is of course one of the important parameters when you want to design a robotic hand in general, so specifically for prosthetics, but in general for robotic hands, you need to set a certain weight. Of course, the Mia hand and the Shadow hand are very different devices, no? So three motors versus 20 plus motors. This comes with a set of different specifications and complexities of the device.

For us, the motor is one of the heaviest components that is inside of the hand. So one way to reduce the weight of the device was to limit the amount of motors that were inside of the device. So what does this mean? Basically, in this way you are setting the limits for the use cases in which you can use your device. Okay, so if your device weighs five hundred grams, it means that you can integrate it also in robotic arms that can carry up to three kilograms. So you will lose half a kilo for the end effector, but you still have two point five kilos for the application. You can use it in prosthetics. On the other side, if your device weighs five kilos, four kilos, you need at least a robotic arm that has a payload of ten kilograms, because if you use a robotic arm with a five kilogram payload, you basically don’t have any additional payload for the task that you need to do.

So it really sets the basis for the set of applications that you need to perform, that you can perform with your device.

Michelle Sun: Speaking of applications, so would you say that the Shadow hand is designed for a different set of applications versus a Mia hand? And what would those two categories be?

Francesco Clemente: Yes. So we designed the Mia hand specifically for prosthetics, and then we reengineered the device, modified the device in order to have also an industrial version. So we have two versions of the device, again, one for prosthetic users and one for industrial settings, so for automating processes in the industry. The Shadow hand is more of a research tool. So far as my knowledge goes, I’ve never seen that robotic hand used in an industrial setting. Of course it’s not usable in a prosthesis, as a prosthesis, but it really targets research settings, where you have capabilities to control the device that are, let’s say, not normal, I would say, because the device is very difficult to use, because you have to control 20 degrees of freedom all at once. Of course it’s becoming more and more simple thanks to AI. So AI is simplifying the use of more complex tools also in more standard settings, I would say. But as far as I know, that amount of complexity is still something that is used more in research settings.

Michelle Sun: That makes sense. And we’re getting to the audience favorite term, degrees of freedom. So everyone is obsessed about how many degrees of freedom, like our human arm has, a hand has, and the tool that you gave me last time was active versus passive degree of freedom. Tell me more about that, and how to read a spec sheet when it says, hey, there’s twenty degrees of freedom on this hand. How many motors are actually there? How many joints are just following along with the motor? Walk me through this vocabulary set of degrees of freedom.

Francesco Clemente: Yes, there is a little bit of confusion there, because degrees of freedom is a general term and then everyone is using it in a bit different way. One way to look at it is using degrees of freedom versus degrees of actuation, where degrees of freedom is number of joints and degrees of actuation is number of motors, or, as you said, passive and active degrees of freedom. So passive ones are the ones that are not controlled directly by a motor, while active ones do. This is the difference between the two.

So why do we have this difference? Because, for instance, if we take a robotic arm, we generally have six joints, six motors, one joint, one motor, and it’s very simple, let’s say. On the other side, in robotic hands, since several years, researchers have been exploring other ways to develop robotic hands, and one way is through underactuation. So underactuation means that the number of joints is larger than the number of motors that drive those joints. And the main reason for this is exactly to reduce the amount of complexity of the device, in some way also to increase the performance. Okay, because one way is the number of movements that you can do, but then you again have grip force, speed, weight, and all of the other performance metrics that you have to compare with and also kind of balance, let’s say.

In this case, what happens if you have less motors than joints, it means that these joints have to be coupled somehow together. Okay, so for instance, in Mia we have three motors that are used to actuate the flexion extension of the thumb, the flexion extension of the index finger, and then the last three fingers are coupled together. So you have three joints, but basically only one motor that controls those joints. And this limits the complexity of the device. This is one way, relatively simple way, to control different joints with a motor, basically with a rigid mechanical transmission.

On the other side, in Azzurra, for example, we have one motor per finger, but one finger has two joints. So we have one tendon that basically wraps around these two joints per finger, and then is used to control both joints simultaneously. This underactuation allows the finger to also wrap around the object and provide some flexibility. So rather than having a rigid transmission between these two joints, we have the tendon and springs inside. This basically allows for the proximal phalanx, in case the proximal phalanx encounters an object, the second phalanx can still continue to flex and adapt to the shape of the object. This flexibility is actually called underactuation and is, let’s say, a smart mechanical system that allows you to still have some adaptability of the shape of the hand to the object without having to actively control all of the degrees of freedom, so all of the joints independently.

Michelle Sun: Got it, got it. So there’s underactuated, when it comes to underactuated hands, then that means that the degrees of freedom is more than the number of motors, the actuators. Which sounds great when it comes to like less overheating, less weight, and less complexity. Is Azzurra also underactuated?

Francesco Clemente: Yes. Because it has eleven degrees of freedom and five motors.

Michelle Sun: I see. And then Mia hand, how many degrees of freedom, and three motors?

Francesco Clemente: That’s six degrees of freedom.

Michelle Sun: Six. Okay. And in theory that sounds great, right? Why wouldn’t all hands do that? Like what is the reason for people to wanna have, is there something like overactuated hands, or, you know, just what’s the opposite of an underactuated hand?

Francesco Clemente: The opposite is, let’s say, fully actuated. So for each joint you have one motor. And this, of course, is the simplest way to develop a robot. So for each joint you have an actuator, and you also have this direct driving, you know, let’s say. The problem, let’s say, with this, and specifically with this setup in a robotic hand, is the compromise that you have to make while choosing the motor and while choosing the performance, let’s say, of your hand, as we said before. You basically have two choices. You end up either having very small motors inside of the joints, because the joints and the fingers are very small, or you need to use some kind of mechanical transmission, like tendons, to have a remote actuation system and then transfer motion. Of course, this allows you to use larger motors, but the device becomes relatively bulky and heavy.

Michelle Sun: And so for Mia hand you use tendons that connect the motors to make it more degrees of freedom than the motors?

Francesco Clemente: No, no, for Mia hand we use rigid mechanical transmission, so linkage.

Michelle Sun: Azzurra hand?

Francesco Clemente: Yeah, that’s the third, yeah.

Michelle Sun: Azzurra, you use tendons?

Francesco Clemente: Yes, correct.

Michelle Sun: Yep. Yeah, makes sense. And you told me that a lot of these motors run five, ten minutes before they need a cooldown. How does a buyer look at the spec, is it possible to look at the spec sheet and know about this mechanism, like the cooldown performance? Is that something that people just don’t publish on the spec sheets?

Francesco Clemente: Yeah, this is very tricky. The problem is that it’s difficult to know from a spec sheet. The reason is that you don’t usually publish heating performance, dissipation performance, of your robotic system. Okay, this is linked to the fact that motors are designed with several parameters. They can run continuously at a specific, when they produce a specific torque. So if you provide a certain nominal current, this current can be provided continuously, and the torque that is generated can be generated continuously. So the motor can run without any stop or without any problem.

The fact is that electrical motors can be provided also with more current than this nominal one. The maximum current that you can provide is called stall current. So when the motor is not moving, you can provide the maximum amount of current, and the motor can generate the maximum amount of torque. Of course, you want to have this torque, because it’s much larger, so two to twenty times larger than the nominal one. So in order to have meaningful performance, you need to exploit these larger currents and torques, but the motor cannot handle these currents for a long amount of time. So this is the reason why you have to stop using it at one hundred percent after some time, because the motor itself cannot dissipate all of that heat that is generated.

So this is something that is well known in the field, and this is the reason why robotic arms are so big, you know, relatively to grippers, because the motors are, let’s say, oversized with respect to the torque that they would actually generate. So they are used at, let’s say, thirty percent maximum performance in order to avoid these issues and be able to run the robotic arm continuously. On the other side, robotic hands are very challenging at the moment. So we don’t really have technologies that allow us to use robotic hands at 30%, 20% of their maximum performance. So it really hits with these heating issues that come up all the time, I would say.

So for a customer it’s very difficult to know this, and I think that you really learn this by experience. So as a researcher that was working with these devices for several years, I’ve seen a lot of these hands, in general also grippers, more advanced grippers with very small joints, run into heating issues, and this is something that we were trying to design for during the years, but it’s something that’s not given for granted.

Michelle Sun: If you were designing the hand for a humanoid that ships ten thousand units or more in a couple of years, how many motors does it have?

Francesco Clemente: That’s a very difficult question. So the question of scalability is an important one. Again, we think that a system, an engineering system, in order to be optimal for the task, has to be sized for that task. I mean, you can design a robotic hand that has 20 motors that of course can perform a lot of movements, but it’s very difficult to scale from an industrial perspective. Okay, so our idea also when developing the Mia hand, and also with new devices that we are developing, is to study grasp taxonomies, because there are several research studies in the field that study grasp taxonomies. We basically study the shape of the human hand while doing tasks, and say, okay, we have the cylindrical grip, the pinch grip, the lateral grip, that are basically grips that are used more often or less often during activities of daily living.

And try to engineer a system that basically covers most of the grasps that are used in doing the activities of daily living. So in this way, you design a device that can perform, let’s say, 90% of the activities of daily living, reducing cost and complexity by 80%, I would say. So I really think that the Mia hand has some limitations, so the fact that you only have three motors, also the fact that these motors can be used all for the thumb position, you only have two positions for the thumb position, so whether it’s opposed or open, you cannot control it actively. It’s a little bit of a limitation for some of the tasks that are required, but probably five motors are a number that I would bet for a robotic hand that is cheap enough to be scaled also from a manufacturing point of view and also a robustness point of view.

Michelle Sun: Interesting. So Mia hand has three motors. Where would the two additional motors be added?

Francesco Clemente: Okay, one could be, of course, on the thumb opposition, because at the moment, I said before that the Mia hand has three motors, one on the flexion of the thumb, flexion of the index, and then the last three fingers are coupled together. Actually the motor that moves the flexion of the index is also coupled with opposition of the thumb, so that with a single motor and a special patented mechanism you can perform the cylindrical grip, but also you can oppose, you can open the thumb, so that you can perform also the lateral grasp. So this movement is coupled with this movement. So when the index finger extends completely, you can move this other joint, let’s say. We would probably decouple those in order to be able to add these independent ones, so that you have a little bit more flexibility and also have a little bit of in-hand manipulation capability.

Probably the other one I would still put in abduction rather than flexion, because there are also studies in the research literature that basically say that having a hand that is able to have abduction movements and two to three degrees of freedom in the flexion of the fingers is equivalent to having a twenty degrees of freedom hand without abduction. So abduction is very important, because it allows the fingers to align, in order to have, so the thumb with the long fingers to align, in order to have a much more stable grasp.

Michelle Sun: I like that. It’s a live design session right here. And you highlight a really important point when it comes to what is the optimal performance level that is also easy to make industrial, on an industrial setting, in mass production, and balancing all these different factors, not just on the spec sheet, like what it looks nice when it comes to the numbers of degrees of freedom. And there’s been a lot of talk about tactile skin, tactile sensors. What are your thoughts about that? They’re not easy to make. They’re very technical and they can be quite bulky as well, depending on what type of tactile sensor that is. Is tactile sensing a prerequisite for these dexterous hands, or do you see that as a luxury?

Francesco Clemente: Probably the best answer from my side is that I don’t really know. I don’t know how much we know about this. So there is a lot of activity going on that demands tactile sensing and very rich tactile information that you use to do dexterous in-hand manipulation. And this, of course, is a specific, let’s say, activity in the manipulation, and in trying to get to something that resembles the human hand. How much this is needed for the tasks that we have to do in, let’s say, an industrial setting, I don’t know. But of course I understand the point of people that want to go deeper into that.

In our devices, we try to have also a balanced approach in this way. We have different amounts of sensors. Some of them are simpler ones. These are basically load cells that are inside of the fingers. So the fact of having the sensors inside of the fingers and not on the fingertip, on the external part, makes them more robust, because not being at the interface with the object, the sensor does not wear out during grasping, so the sensor lifespan is increased. On the other side, it’s more difficult to get position information, so spatial information about the location of the contact point between the finger and the object. For this, you really need a matrix of tactile sensors that are on the fingertip, and this is something that you cannot really go around.

We have partnered with Touchlab, which is a UK company that develops some of the sensors that tie inside of our hands, and that work very well. For instance, we have developed the Mia hand with these sensors, with a matrix of tactile sensors on the thumb, and this allows for richer, let’s say, information to come out from the hand. From a deployment point of view, there is a lot of discussion about the fact that if these sensors are required in, let’s say, a teaching phase of the robot, and then when you deploy the robot, maybe you need simpler sensors, or no sensors, in order to perform the task in a continuous way. But this is something that is still to be discussed and understood.

Michelle Sun: Interesting. Did you mention that the Mia hand doesn’t have tactile sensor on the fingertip, but on other parts of the hand? And did you mention that there’s also sensor on the thumb?

Francesco Clemente: Yes, correct. So the Mia hand has load cells in the thumb, index, and middle finger, at the base of the finger. And then this part, or, yes, at the proximal part of the fingers, I would say, and this allows for the sensors to be sensitive on the whole finger. So it’s not important where you’re touching, the sensor will still pick up deformation. But as I said before, on the other side, you don’t know the location of the contact point.

Michelle Sun: Very cool. If you take the Mia hand apart on a table with all the parts that go into it, motors, gears, sensors, machine parts, can you walk me through, let’s say, the three actuators, like how many, what’s the percentage of the bill of materials that it costs, and what is the most expensive line item in the hand?

Francesco Clemente: Yeah, the most expensive line item, I will say that the motors and the force sensors, so those are the two most expensive ones. Then the mechanical components are, of course, accounting for probably 70% of the cost. So most of the cost is mechanical transmission and the frames of the device. And then you have the electronic components, the PCBs, those ones really are much cheaper to do also at a larger scale, so they don’t really account for a lot of cost in the bill of materials. But of course the motors and advanced sensors, for instance the ones that are a matrix of tactile sensors, I would say, are the most expensive ones.

Michelle Sun: For the semi percent of the transmission related costs, how much of that would be the three actuators and the sensors?

Francesco Clemente: On top of my mind I would say that the motors account for maybe thirty five percent, more or less, of the total cost. Well, we don’t have a lot of motors, so for hands that have a lot of motors this number goes up a lot.

Michelle Sun: I mean, thirty five percent is actually not that out of the range when it comes to like the numbers that I’ve been seeing, like forty to sixty percent of a humanoid is actuators. I’m assuming another bulk of it would maybe, another twenty percent is the touch sensors. Is that how you’re seeing it?

Francesco Clemente: Yes, more or less.

Michelle Sun: And where do these motors and gearboxes come from? You mentioned that you work with Touchlab for the touch sensors. Do you also work with a European supply chain when it comes to these hand scale actuators, or are you sourcing from Shenzhen?

Francesco Clemente: No, we use suppliers from Europe for the motors, because we basically could not find a replacement for these motors that was keeping pace with the performance. So we never switched to other solutions.

Michelle Sun: I see. So the European supplier that you have is able to give you off-the-shelf actuators that can be replaced pretty quickly. Or did you have to do modification in-house to fit the motor for the Mia hand?

Francesco Clemente: No, no, we use off-the-shelf actuators, actually. So usually the motor suppliers have a catalog of motors and gearheads that you can combine together. So basically what we do during design is combine motors plus gearhead, in order to have the best combination to get the torque and the speed that is needed for the application. We have a little bit of customization that we sometimes discuss with the suppliers, but it usually only gets down to the size of the output shaft, or the shape of the output shaft, in order to be able to connect more easily the motor to the rest of the mechanical transmission.

Michelle Sun: Who actually buys a Mia hand today? Who are your customers? How do they split between prosthetics, industrial sales, humanoids, and research labs?

Francesco Clemente: Yes. So at the moment we started selling Mia hand as a research device as well. So before it was also CE marked. Most of our customers are still in the research space. We CE marked Mia hand at the beginning of twenty twenty five. So at the moment the hand is CE marked both as a prosthesis and as an industrial gripper. I would say that we still have at the moment more customers from the prosthetic side, and the industrial side is coming out later. I would say fifty percent research, thirty five prosthetics, and fifteen percent industrial. We are now pushing a lot on the industrial side, because we believe that’s an early adoption market in that case, because industry is not really used to using robotic hands, but they’re accustomed with grippers, but we see rising demand for more flexibility. So we want to push on that.

Michelle Sun: Do you have any visibility in terms of what kind of industrial task your customers use Mia hand for? And what are the customer profiles of these industrial customers? If you were to point at one particular deployment task out in the world and say, hey, this is what actually pays the bills, and what you see as the biggest growth area for Mia hands, which one would it be?

Francesco Clemente: Yes, we are very excited about two use cases. One is testing of human machine interfaces. So all of those objects and goods that have a human machine interface that is designed to be used with the human hand, so by humans, like the interior of cars, so in the automotive sector, at the moment you have at the end of the line a person that goes inside of the car and tests the interior of the car. This can really be automated with a device like Mia hand, that is lightweight and can fit together with a robotic arm inside of the car, and be used to test all of these components. But also like medical devices, like ultrasound machines, ekographs, that have keyboards and mice, that have these interfaces that are relatively advanced, not just simple buttons, but big buttons, levers, a mouse, and also a touch screen. So here we see that Mia hand can be used because of its flexibility, and you cannot really use a simple push-button gripper, let’s say.

And the other one is HoReCa, so hotelier, restaurants, and catering. So serving food and preparing food. In that case, we think that both the flexibility of the device and also the aesthetics of the device are very important, because you have to interact with very different objects while serving food and preparing food. On the other side, you are also in contact with the general public, so this is something, so also aesthetics is very important. We have worked with partners in the past in order to develop robotic cells that prepare ice cream or serve you food, and this is also something that we look for.

Michelle Sun: With Oversonic, the Italian humanoid customer, they use your hand on the robots. Walk me through how that actually happened, from the first email to the first production order. How long does it take, and how many units do they eventually buy?

Francesco Clemente: Yes, Oversonic is an Italian company that develops humanoid robots. So there are not a lot of Italian companies that develop humanoid robots, and there are not a lot of Italian companies that develop robotic hands. So it was relatively simple for us to get in contact. I think that one of the first contacts we got was at an industrial fair in Bologna, so we met them there, and we started discussing possible collaborations, because they found the development of the robotic hand to be very difficult, and we already had a device that could be fitted to the robot. So they were happy to discuss collaboration with us. So they have, at the moment, different use cases.

The one they are exploring the robotic hand, so Mia hand, the most, is with their medical version of the robot. So they have a medical version of Robee, that is the name of their robot, that is used inside of hospitals to support rehabilitation therapies inside of the hospital. So also in that case you need, it’s not an industrial setting, but you need flexibility in order to interact with patients, interact with different objects, rehabilitation devices, let’s say. They were looking for somebody that was resembling the shape of the human hand and that would provide the flexibility in the grasping and manipulation that Mia hand could provide. In this way basically we found this use case that was interesting for them, and they were basically deploying some of their robots in their first pilots inside of this use case.

Michelle Sun: Amazing. Seems like this medical and hospital use is a recurring theme that you mentioned, about like ultrasound machines, and then with Oversonic as well. If we zoom out on the whole dexterous hand market, Psionic also works on prosthetic hands like yourself, and now their Ability Hand sells into robotics, also around like fifteen to twenty thousand dollars. Inspire Robotics got a lot of humanoids using them. And then Shadow, you mentioned, sells mostly into research institutions. And then there’s also Robotiq in Canada that owns the industrial grippers segment. Who do you see as your competitor, and who’s in a completely different business that just looks similar on the form factor?

Francesco Clemente: From an industrial point of view, I think that from the people that you’ve mentioned, let’s say that the Shadow robotic hand is a little bit different with respect to the other ones, because as far as I know, as far as what I see, they’re very much focused on research only. So they’re not selling their device to be used in industry and in prosthetics. So that’s relatively different. And also the hand is very much different from the devices that we develop, so there is also a matter of really different specifications between the two.

Of course, the Ability Hand is the closest one to us, so that’s, of course, a competitor for us. I think that our devices are a little bit different, with different specific strengths and weaknesses, but definitely we are playing in the same space. Robotiq, that you mentioned, is, of course, one of the largest gripper manufacturers in the world. They focus a lot on standard industrial grippers. So again, Robotiq or Schunk or these kind of people, so they have a very strong market in the industrial settings, let’s say. What we would like to do is complement their offering with something that is more flexible. Okay, so our competitors, meaning that probably the clients in the industrial settings that they have and the clients that we have are the same, but we are offering something different.

So let’s say that the device that we offer is complementary with respect to the standard gripper, because the standard gripper is something that is great if you need no flexibility, but you need very high performance, so very fast, very strong, very repeatable, and that’s it. But you only can grip, let’s say, a single object. While we are providing something that is more flexible, still industrial grade, but lower in performance. So we are not able to match the grip force and the speed of a gripper, because that’s a very specialized tool.

Michelle Sun: And it seems like it’s a different performance and different use case as well. So when it comes to grippers, it’s about picking one thing to another place mostly, versus when you mentioned about the use cases that you see, the two main ones with HoReCa and also the human interface, is more about more delicate objects, or you touch some screens or moving a joystick in the ultrasound machine. So is that how you think about it? Like the gripping part, like pick and place mechanism, belongs to the gripper form factor, versus the Mia hand is able to do something that is more about, for example, touching a screen or testing the automotive interior, that needs more of a grip, finger based grip sensation, like action.

Francesco Clemente: Yes, this is one way to think about it. So it’s the specificity of the gripper, so the fact that it’s very much designed for pick and place and for moving objects quickly, while Mia hand, of course, it’s more flexible, so that you can do different things in a specific use case. But not only that, the flexibility comes down to the flexibility of the device. So on one side the gripper is usually designed to be used, so from the start to the end of its lifetime, for the same activity. So you design the fingers of the gripper in order to pick a specific object, because you want it to grip that thing and do that without changing it. While on the other side, a hand, the fact that it’s a hand, gives the device more flexibility. So this device is designed in order to be also relocated to different tasks, so not doing the same thing over and over again. But let’s say, okay, today I need to do these tasks, in two months I need to do something else, and this is something that is enabled by the flexibility of the device.

Michelle Sun: My last question is, when you think about owning that market when it comes to complementing the industrial grippers and supplying to these industrial use cases at scale, what are the biggest blockers when it comes to the manufacturing side? Do you see yourself needing to find, like, a different supplier to build a Mia hand at tens of thousands or hundreds of thousands at volume? Walk me through that future.

Francesco Clemente: Yes, of course, this is something that we are looking into at the moment. So at the moment, Prensilia is basically a design firm. So we have a strong engineering team that designs all of the components, and we outsource the production. And on the other side, we get all of the components, assemble, and do the quality tests in-house. And this allows us to ensure a good quality of the device. Of course, this can run up to a certain scale, let’s say hundreds of hands per year. But if we would have to go to thousands of devices per year, then we need to probably change approach. We are already looking into this in two ways, let’s say: redesigning some of the components in order to use manufacturing processes that allow us to scale production from a manufacturing point of view, but also partnering with partners that help us, let’s say, with assembling part of the device, in order to be able also to scale manufacturing capabilities on our side. So we think that we are not gonna solve this by ourselves, but we are now partnering with strong partners that are specialized in manufacturing, in order to be able to scale production and meet the demand that seems to be needed for the next years.

Michelle Sun: Awesome. Well, thank you so much, Francesco. This is amazing, to chat with you about everything from the design and the mechanism that goes into dexterous hands, but also the market that you’re seeing on the ground. Thank you so much for joining today.

Francesco Clemente: Thank you. I mean, it was a pleasure.

Subscribe now

Subscribe now

Subscribe now

Robot models are not yet like LLMs. We do not use them daily for work or at home. That is why the releases are harder to visualize.

In 2026, every week, or multiple times a week, we see new robot foundation model releases. Many of them claim a GPT-3 moment for robotics. They are all impressive. At the same time, terms like “in-context learning” and “scaling laws” can feel abstract to many (myself included!).

When I read each release, I want to know these four things. What does it actually do, in a humanly relatable way? How does it compare with previous models of the same company and other models out there? What does it unlock? And what’s next?

This format is inspired by the WIRED series that explains complex ideas through different levels of detail. Two of my favorites are its pieces on sleep and machine learning.

In this piece, I share my mental model for answering those questions, one release at a time. Here’s the structure:

• Explain it to a 5 year old

• Explain it to a smart 12 year old

• Explain it to an undergraduate student

• Explain it to an expert peer

• Which bottleneck is it solving

• What it is trying to solve next

I organize by release date, most recent first. I will make this a series as new models ship, so we can compare them over time. If you are new here, subscribe to get the future pieces in your inbox.

For ease of referring to the robot, let’s call it Otto, like autonomous :) These models are Otto’s brain. Each section below gives Otto a different one. “Showing Otto” means giving the model a short recording of someone performing the task. It is closer to uploading an example into a chatbot than to retraining a robot from scratch.

We will go through Skild AI’s S1, Generalist AI’s GEN-1.5, Dyna Robotics’ Dyna-2, Sunday Robotics’ ACT-2, Xiaomi’s Xiaomi-Robotics-1, Robbyant’s LingBot-VA 2.0, and Physical Intelligence’s π0.7.

NVIDIA’s GR00T N1.7 (July 7) expanded the underlying stack rather than demonstrating a new way for robots to learn. We omit it here to focus strictly on task-learning breakthroughs.

Every number in this piece is as reported by the companies themselves.

As always, comments and feedback are greatly appreciated.

Skild AI, S1

Release date: August 25, 2026

To a 5 year old. Dad records a video of himself making pancakes. Otto watches the recording once. Then Otto tries to make pancakes on its own.

To a smart 12 year old. Previously, teaching Otto a new task meant collecting hundreds of examples. Then running a training session for that one task. With S1, Otto learns to make pancakes from one video of Dad doing it. No upgrade to S1.5 or S2 needed. The same model handles each new task, one video at a time.

To an undergraduate student. In the GPT-2 era, each new task needed its own fine-tuning run. Want a model to sort movie reviews into positive and negative? You gather labeled reviews. Then you train. GPT-3’s breakthrough was that examples could go in the prompt itself. Give it “sea otter: loutre de mer” and “whale: baleine.” Then ask it to translate a new animal. No training run. S1 does the physical version of that. The video demonstration is the prompt. The robot executes it with the same model, weights unchanged.

To an expert peer. Skild calls this in-context learning for robot policies. S1 conditions on a video demonstration at deployment time. It does not update its weights. On unseen tasks, Skild reports 66% cumulative per-step success, versus 9% for its language-conditioned VLA baseline, after pre-training on 100,000 hours. One in-context video performed like about 380 post-training episodes. Skild says those episodes would require 50 to 100 hours of teleoperation.

Bottleneck it’s solving. Teaching a new task is expensive. It used to mean a technician’s week of robot operation. S1 changes the unit of teaching to a single recorded example, weights unchanged.

The GPT-3 parallel. GPT-3 made a new language task cheaper to specify. Users could put examples in the prompt. Before that, only machine learning engineers could train a new model. After it, anyone could teach one. GPT-4 and the models that followed did not reinvent prompting. They made it reliable enough to build products on. S1 is reaching for the same shift in robotics. Deploying a new task takes one demonstration instead of a retraining run. Over time, the same model accumulates a long task list.

What it is trying to solve next. One caveat on the numbers first. Skild grades tasks step by step. The score counts the steps Otto completed. If Otto stalls, a human can nudge it forward so the later steps still get scored. So 66% is the share of steps completed. It is not the share of tasks Otto finished alone.

The open questions:

• Can one demonstration reliably produce a clean, unassisted run?

• Does performance hold across unfamiliar objects, layouts, and robot bodies?

• Can a one-off demonstration become a robust production system? The system is more than the robot. It includes how demos get recorded and checked, which tasks get accepted, and how failures get caught. Deployment needs a whole system to work.

Generalist AI, GEN-1.5

Release date: August 19, 2026

To a 5 year old. Show Otto a three-second video of Dad twisting open a jar. Otto tries to twist open the jar.

To a smart 12 year old. Otto can start opening jars from one very short example. Give it a few more minutes of Dad doing the same thing, plus a quick update of the brain. Jar-opening success goes from about 59% to 83%.

To an undergraduate student. A chatbot can pick up a pattern from examples you paste into the conversation. Show it two invoice lines coded the right way. It codes the next one the same way. No retraining. That behavior has now shown up in robotics. GEN-1.5 can learn physical skills on the spot, without being trained on that skill explicitly. It offers two modes. One short physical example for immediate behavior. Or a few minutes of data and a handful of updates for a small, lasting adjustment.

To an expert peer. Generalist describes GEN-1.5 as in-context learning with optional few-step fine-tuning. One 3 to 12 second demonstration produced 59% average success across 10 tasks, with a reported plus-or-minus 10 percentage-point spread. One gradient step on one minute of data reached 66.5% on a held-out task. Ten steps on about five minutes, or roughly 50 demonstrations, reached 83%, with a reported plus-or-minus 9 points. The pre-training set contains 270,000 hours of UMI data and grows by 10,000 hours per week.

Bottleneck it’s solving. Teaching a new task is expensive, and this cuts the bill. Task-specific data and compute drop to minutes and a handful of updates.

What it is trying to solve next. These are short, simple tasks: jars, zippers, pouches. The open question is whether the same pattern extends to long-horizon work.

Dyna Robotics, Dyna-2

Release date: August 10, 2026

To a 5 year old. Dad wears a head camera and records himself chopping tomatoes for dinner. Otto watches many videos like that. Then Otto tries chopping with its own robot hands.

To a smart 12 year old. Dyna teaches Otto how the world changes. When a knife cuts a tomato, the tomato splits into slices. Otto learned all of this from watching videos of people. It can translate human hand movements into robot hand movements.

To an undergraduate student. Most robot models are VLAs, vision-language-action models. They look at the scene, read the instruction, and output an action. Dyna-2 is a WAM, a world-action model. It also predicts how the scene will change next. The actions come out of that same model. Understanding how the world evolves helps Otto pick a better next move, especially somewhere it has never been.

To an expert peer. Dyna calls Dyna-2 a world-action model, not a VLA. It uses a video-diffusion backbone with a mixture of transformers, flow matching, and DiT layers with causal masking. Dyna reports more than 1 million hours of egocentric human video for pre-training. Its evaluation covers 14 benchmark tasks across 3 embodiments, with 39 tasks in total. The reported customer-site pass rate rose from 46% with DYNA-1 to 87% with Dyna-2.

This matters for a second reason. Dyna’s pre-training data is ordinary human video captured on head-mounted cameras. No robot gloves. No teleoperation rigs. That source is cheaper and far more abundant than robot data. In Dyna’s experiments, the video prediction is what made human-to-robot transfer improve as data scaled. Their WAM reached 1.55x the success rate of their previous VLA.

Bottleneck it’s solving. Robot data is scarce, and teleoperated demonstrations are the expensive kind. Dyna learns world dynamics from large-scale human video instead. If that reliably transfers to robots, we need far less of the scarce stuff.

What it is trying to solve next.

• Does human video keep improving robot performance on very different robot bodies, not just human-shaped ones?

• How much robot-specific data is still required after a million hours of human video?

• Can the 87% customer-site result be reproduced by independent operators and across more sites?

Sunday Robotics, ACT-2

Release date: July 17, 2026

To a 5 year old. Otto folds laundry. It folds in homes it has never visited. It folds clothes it has never seen. Even when clothes fall on the floor, Otto picks them up and keeps folding.

To a smart 12 year old. Otto practiced folding laundry in Sunday’s lab. It practiced until it could fold 99 times out of 100. Usually that kind of practice only works in the practice room. A new home breaks it. ACT-2 is the exception. Otto folds laundry in homes it has never visited, with clothes it has never seen.

To an undergraduate student. ChatGPT gets better at your style of writing after a few corrections. It carries the lesson into the next conversation instead of forgetting it. ACT-2 does the physical version. One fine-tuning example teaches a new behavior. The improvement generalizes to unseen homes instead of sticking only to the practice room.

To an expert peer. Sunday uses “solve” for reliable performance across a declared scope at a stated adaptation cost. For ACT-2, the declared task is laundry and the stated per-home adaptation cost is zero. Sunday reports 99 successful folds out of 100 in its lab evaluation. The claim to test is transfer: the same behavior is reported in unseen homes and with unseen clothing, without collecting new data or fine-tuning in each home.

Bottleneck it’s solving. Skills usually break in new environments, so each new home would need its own data collection and fine-tuning. ACT-2 needs neither: no data collection or fine-tuning per home, with training data from people wearing robot gloves.

What it is trying to solve next.

• Laundry is one task. Vacuuming, toy tidying, zippers, and coffee are still in training.

• Does the reliability survive months in a real family home, not just evaluation visits?

Xiaomi, Xiaomi-Robotics-1

Release date: July 16, 2026

To a 5 year old. Dad loads the laundry into the washing machine. He uses a handheld robot hand. A small camera on the gripper records everything. Otto watches many recordings like that. Then Otto loads laundry with its own hands.

To a smart 12 year old. Many people recorded themselves doing chores. They used a small handheld gripper with a camera. Then the robot practiced the same chores with its own body. The recordings are not tied to one robot. So the same model can learn the tasks on different robot bodies. In Xiaomi’s tests, under 10 hours of examples per task reached a 75% average success rate.

To an undergraduate student. This is similar to training ChatGPT on a large amount of general material. Then giving it a smaller set of examples in the format it will actually be used in. Xiaomi pre-trained on broad human manipulation data. Then a smaller amount of robot data taught the model to follow language instructions on real robots.

To an expert peer. Xiaomi-Robotics-1 uses embodiment-free UMI trajectories for pre-training, then real-robot data for post-training. Xiaomi reports 100,000 hours of UMI data across more than 1,700 scenarios, followed by more than 7,200 hours of real-robot data from homes and cross-embodiment datasets. Under 10 hours of demonstrations per task produced 75% average success, versus 40% for the baseline. Under 40 hours produced 85%, versus 53% for the baseline.

Bottleneck it’s solving. Robot data is scarce, so Xiaomi splits collection in two: pre-training on portable-gripper data anyone can record, then post-training on real robots. It also publishes how much demonstration data each task took.

What it is trying to solve next.

• Do these numbers hold outside Xiaomi’s own tests?

• Does the handheld-gripper-to-robot transfer hold across more robot bodies and harder tasks?

• What does a demonstration hour actually cost? Recording is one part. Annotation, setup, and supervision are the rest.

Share Core Matter

Robbyant (Ant Group), LingBot-VA 2.0

Release date: July 9, 2026 (arXiv, revised July 16; peer-reviewed at RSS 2026)

To a 5 year old. Otto carries a glass of water across an obstacle course. While it takes one step, it is already getting ready for the next one. It does not stop to think between steps.

To a smart 12 year old. If Otto stopped after every movement to think about the next one, it would be slow and jerky. The water in the glass would slosh out. It would lose the race. LingBot predicts what will happen next while Otto is still moving. Then it checks what actually happened and adjusts.

To an undergraduate student. There is no clean chatbot parallel here. ChatGPT generates one token after another. The next token waits for the last. LingBot-VA 2.0 predicts and executes at the same time. It predicts future video states and actions in parallel with what the robot is doing.

Picture the AI autocomplete in email apps. It drafts the rest of your sentence from context while you are still typing. LingBot does that, except the sentence is a physical movement. Otto keeps moving while it drafts.

To an expert peer. Robbyant describes LingBot-VA 2.0 as an autoregressive video-action model with closed-loop control. It uses a semantic visual-action tokenizer, a shared observation-action latent space, a sparse mixture-of-experts backbone, and asynchronous inference. Future visual latents are predicted while the current action executes. The model re-grounds each rollout on the latest observation. The paper is peer-reviewed at RSS 2026 and does not publish a task-cost number.

Bottleneck it’s solving. Robots pause between thinking and acting. Plan too slowly, and the robot is slow and jerky. Act too fast on stale predictions, and the robot keeps moving after the environment changes, eg. a chair gets shifted, something rolls into its path. LingBot predicts and acts asynchronously in a closed loop, so Otto keeps moving smoothly while staying responsive. Real environments do not pause while the robot thinks.

What it is trying to solve next.

• Can the robot keep making accurate next-move predictions at speed, over long tasks?

• Can it react when objects around it move unexpectedly?

Physical Intelligence, π0.7

Release date: April 16, 2026

To a 5 year old. Otto has one toolbox. Not a separate box for pancakes, one for laundry, and one for making coffee. After dinner, Otto cleans up the kitchen. It picks different tools from the same box for each job.

To a smart 12 year old. Before, robots often had a separate specialist model for each task. One for opening the drawer. Another for picking things up. Another for wiping the counter. With π0.7, Otto uses one brain for the whole kitchen cleanup. It matches or beats the separate models.

To an undergraduate student. Instead of having one AI for translation, another for trip planning, and another for financial analysis, you use Claude for all three. π0.7 attempts the physical equivalent. One model controls different robots across different tasks. The analogy is not perfect. Claude produces text in one interface. π0.7 still has to translate instructions into safe movements through a physical body. That translation is called grounding. It connects words like “wipe the counter” to the actual forces, paths, and motions that work for this robot, in this kitchen.

To an expert peer. Physical Intelligence frames π0.7 as “a step-change in generalization.” It uses multimodal prompts with language instructions, world-model-generated visual subgoals, speed and quality metadata, and control-modality labels. One π0.7 model matched or beat π*0.6 task-specific RL specialists in the company’s controlled experiments. The post does not state the total training-data scale or the cost of adapting to a new task. That missing number matters when comparing it with the models that publish demonstration hours.

Bottleneck it’s solving. When every task needs a separate model, deployment becomes a model-management problem. Different weights, interfaces, update cycles, and failure modes to track. π0.7 is one general model across tasks and robots, which simplifies that stack. It also shows a robot attempting tasks it was never taught, like operating a kitchen gadget it has not seen before.

What it is trying to solve next.

• Can it do genuinely unseen tasks reliably, not just recombinations of familiar skills?

• How well does grounding hold when the scene, the objects, or the robot body change?

• How much does it cost, in data and in compute, to adapt the unified model to a new task?

Where this leaves us

One note on the tool behind several of these models. UMI, the Universal Manipulation Interface, is a portable gripper with a camera. A person carries it while performing a task. It captures the person’s movement and the scene at the same time. The original open-source design costs about $370 in parts. The 3D-printed gripper is $73. The GoPro camera and accessories are $298, per the paper that introduced it. That is cheaper and easier to scale than teleoperating an actual robot. Teleoperation needs expensive hardware, setup time, and skilled operators. Sunday built its own variant: sensorized gloves at about $200 a pair.

In summary:

• Xiaomi-Robotics-1 pre-trains Otto on humans using handheld grippers, then adapts to new tasks with a few hours of real-robot demonstrations

• S1 lets Otto learn from one example

• GEN-1.5 enables Otto to learn in seconds and minutes for quick tasks

• Dyna-2 teaches Otto from large-scale human video by predicting the environment

• ACT-2 lets Otto do a task reliably in homes it has never seen, with no per-home training

• π0.7 lets Otto do many tasks with one general brain, not a collection of specialized brains

• LingBot-VA 2.0 predicts what happens next to help Otto move quickly and smoothly

The releases are moving the frontier in different directions, but they leave the buyer’s main question partly unanswered: what does it cost to make one new task reliable in production? I’ll dig into that in a follow-up piece in the coming weeks.

In this episode we cover:

• Why fresh produce still has no automated harvester

• The labor math underneath agricultural robotics

• Physics that replaces chemicals, and vision that reduces them

• What actually breaks a machine in a field

• The two-year payback every agricultural machine has to clear

• Who manufactures and services these machines

• Turning farm jobs into ag tech jobs

Danny Bernstein on LinkedIn
Reservoir Farms Website
Watch on YouTube

Subscribe now

Every episode, I sit down with the founders, researchers, and operators building physical AI worldwide, tracing the full stack from components and supply chains to what actually deploys. See all published episodes here.

Chapters

• 00:49 Danny Bernstein, Reservoir, and a technical community for agriculture

• 00:49 Silicon Valley builds seven of everything, agriculture waits for one

• 03:15 Driscoll’s, Taylor Farms, Dole, and a $4.8 billion county

• 05:29 Not AI for luxury, and the three gaps worth building into

• 05:29 Precision surgery by robot, fruit picked by hand

• 07:50 400,000 advertised farm jobs and 182 domestic applicants

• 10:11 What actually breaks a machine in a field

• 12:36 Sonoma, and the plan to cover the top ten specialty crops

• 12:36 Washington, tree fruit, and the Cosmic Crisp patent

• 14:58 Two years to return the cost of the machine

• 17:14 Service teams, dealers, and Andros Engineering

• 21:50 Merced College, and building on a teaching farm

• 24:12 YC gives you tokens, Reservoir gives you a tracto

We are performing precision surgeries using robots, but we still hand harvest fruit. And there were over four hundred thousand of these jobs that were posted in twenty twenty five and only a hundred and eighty-two domestic applicants for those positions. These are jobs that Americans are simply don’t want to do sitting where you’re at today, what are some biggest gaps that you’ve seen that you wish more people are building? And not AI for luxury or

AI’s for abundance, but it’s really AI for resiliency. Why combinator gets a million dollars in tokens from Oken AI, we get access to a tractor.

Danny, welcome to the show. Thank you, Michelle. This is super fun. Thanks for coming out to Salinas. Yeah, I know this is a great day to be out here. So we’re out here, Salinas, 24 acres. 40 acres. 40 acres of you know, like different machines running on the field. So take me back to where this idea first started. Yeah, so Reservoir is a technical community for agriculture, and that’s kind of a new thing. And and so what inspired me is that I actually spent close to 20 years in Silicon Valley.

How impactful technical communities were toward accelerating big discoveries. And you’re in San Francisco and there are technical communities kind of everywhere, whether it’s Y Combinator or around UC Berkeley, around Stanford. And I started to study basically how were technologies developed for the physical world where these technical communities may not exist, and we have significant technical opportunities in areas like agriculture or forestry or land management.

But like where are the startups and how do they connect with each other? And that is like the, you know, 30 second version of kind of how we came up with this idea, which is what if we built like the most robust technical community around Ag and what if we did it in the most important growing region, Archieville, in the United States? Tell me more about Salinas, right? How did you pick this part of California? What’s so special about it? Yeah, so I spent close to twenty years in Silicon Valley. I was part of a startup that sold to Google.

And I was at Google for 10 years and I spent a couple of years at Microsoft. And what really defined my time in Silicon Valley was sort of the behavior of how Silicon Valley developed products, which is one company comes out with a thing, and that thing is the new thing. Let’s say it was Zoom or it was quad code or whatever the first thing is. And then every company basically comes up with their version of that thing. And they basically are competing around what is essentially a solve problem.

And then if you flip it around basically and you look at what are the areas that don’t have that kind of activity, where you don’t have seven companies competing to solve one challenge. And agriculture is like that, where we have a bunch of really significant gaps in technology. And we’ll talk about it. But like an example would be automated harvest for fresh produce. Simple idea. Automated harvesters, automated robots for harvest, specifically in fresh produce where the output is more delicate.

We have practically no solutions. So you contrast that with Silicon Valley, where they’re like seven different ways to vibe code or seven different ways to do video calling, or seven different ways to message someone or email someone, right? We have an abundance of solutions we have to choose. You flip that around and you look at an area like agriculture where you don’t have that abundance. So we started calling it technology as resilience. And then if you look at basically agriculture in particular, and you were to hone in on fresh produce.

which is on micronutrients supply, like our most valuable calories are our micronutrients. And that’s fresh produce. And Salinas ends up being really the fresh produce capital of the United States, if not the world. Yeah, so Monterey County is a five billion dollar ag economy and Salinas is really the center point of that. So where we’re sitting right now, we are within driving distance of the headquarters of Driscoll’s, of Taylor Farms, of Dole Vegetables, all these like incredibly

massive vegetable and fruit production companies. So where do you build the technical community around those products? You do it here. So you’re really in the backyard of all these like the center of all these different agricultural powerhouse. When you talk to like companies like Drisle or, you know, even John Deere that is like in the space for a long time, like how do they view like ArcTech and the autonomous

wave that’s coming to this industry. Yeah, I mean they view it like a imperative, like it’s a must happen. It’s not a nice to have, it’s a must have. And it’s because of labor and input related costs. So chemicals, inputs, fertilizer, all the sort of ingredients of farming are becoming more expensive and more regulated. And then the labor of of agriculture is becoming increasingly expensive and scarce. We cannot pack each of those.

And so then they basically view technology as the only way out of those challenges. There isn’t really an alternative to technology. We’re not going to suddenly have millions of people who want to get into farming. It’s not going happen. We’re not going to suddenly have a pipeline of new chemicals that are going to be better than what was previously there because of anything we have more and more sort of chemical resistance in our crops now. And we haven’t had significant novel new solutions in chemistry come into ag in a very long time. It’s like something like two in the last thirty years.

It’s the view of growers, of operators, of agribusinesses, even of communities, that technology is essential. And so it’s viewed as one of the areas of AI that are not AI for luxury or AI sort of for abundance, but it’s really AI for resiliency. And that’s a different story altogether. And you mentioned about how there are so many gaps in

you know, how AI is solving in agriculture. It’s different from like vibe coding, having different so many different ways to play around with that. Tell me more, setting where you’re at today, what are some biggest gaps that you’ve seen that you wish more people are building? Yeah, sure. So I we really put them in three categories. The first is using physics to replace chemicals. The second is using AI or machine learning or vision.

to reduce chemical use, basically more targeted application of chemicals. So chemical replacement, chemical reduction. And then the third one is automating the jobs that nobody wants to do. And we have a lot of really hard jobs in agriculture right now. So we’re in Salinas, it’s about 70 degrees today. But if you drive two hours east to the Central Valley where we’re in the middle of harvest or kind of in the middle of tree fruit harvest, it could be 105 today. We do not have any automated harvest technology in Stonefruit.

We hand harvest peaches, we hand harvest table grapes. All of those sort of incredible fresh produce items require hand harvest still in 2026. we are performing precision surgeries, you know, using robots, but we still hand harvest fruit. Those are the three categories. Physics to replace chemicals, AI, sort of machine learning and machine vision to reduce chemical use and automating the hardest-to-fill jobs. Remember hearing a case where, you know, with these harvests.

being so such a labor intensive and still dependent on like human going onto the field is actually pretty dangerous, right? Like they get heat stroke and, you know, hours and hours under the sun. It’s actually like not safe to work so many hours. And so that’s where machines can really increase our quality of life and also lower the cost of Yeah, these are jobs that Americans simply don’t want to do. There was a the American Farm Bureau performed a study in twenty twenty five. They looked at

The 400,000 jobs that were posted were basically farm workers, and there were only 182 domestic applicants. Wow. So it’s 0.04% of the 400,000 jobs that were posted received a domestic applicant. The reason they know that is because there’s a visa that is our for foreign sort of the the for the farm worker that’s brought in from a place like Mexico to the United States to do that work. It’s called an H-2A visa.

And it’s how we actually perform harvest today is using imported labor that stays in rural communities for ten, eleven months in specific housing called, you know, H-2A housing or farm worker housing. Each individual H-2A job is posted publicly to ensure that there isn’t a domestic applicant for that particular position. And there were over four hundred thousand of these jobs that were posted in twenty twenty five.

And only a hundred and eighty-two domestic applicants for those positions. It’s wild. And so basically you’re saying that even though we have an unemployment in this country and some of our rural communities had nine, ten, eleven percent unemployment, these are jobs that Americans are simply not even applying for. And there are a variety of reasons why. It’s hard work. We also have a gig economy now that if you don’t want to do outdoor work, you can just drive a car and pick up, you know, Doordash.

So there are different ways to make ends meet now that didn’t exist in previous generations. So yeah, people do not want to work on farms. These are hard jobs. And technology is really for markets like the United States or Japan or Australia where there’s a premium on labor and labor is expensive. Technology is the gap. It’s that’s the solve. Yeah, and definitely in not just America, like globally, you’re seeing people abroad sit in front of their computers or on their phones or driving car.

Delivery, there’s definitely this gap that machines can help bridge. You sitting out here seeing all these machines running on the fields, what are the things or conditions that really break a machine? Is that the dust or is that the humidity, vibration, or you know, what are the surprising things that you’ve seen that are you saying yes to all of those? But if we take a step back, the robotics talent, the physical AI talent, if we think about where it’s originating from, it’s mostly from

PhD labs or from computer science schools or from places like Google or Nvidia or whatever, you have individuals that are, you know, in Palo Alto, in San Francisco, in Cambridge Mass, and if they decide tomorrow that they want to start a agricultural robotics company, like where do they begin? And so a lot of what we did to design Reservoir Farms is we actually thought about that persona. So I’ll very often say to the team, let’s think about the roboticist in Palo Alto and what she or he is looking for in a startup space.

And that’s how we design a Reservoir Farm. So we have forty acres behind us. It has leafy greens, so vegetables and strawberries. Right now those forty acres are it’s sort of farmed like an R and D farm or a test farm. So we have a few acres of iceberg, then a few acres of romaine lettuce, then we have five acres of strawberries, which is easily the most expensive to farm, like exponentially more expensive to farm. They it red gold. We can talk about that. And then we have some brassicas, we have broccoli and cauliflower.

So in one farm they can access behind us, you can access several different ultra high value commodities to be able to build your machine. They can get a lot of what those sorts of conditions are. Our particular farm is flat. It’s actually laser leveled. Most of the farms in Salinas Valley are laser leveled. If you go thirty minutes north, Castroville or or Prendale, you’ll see a lot of strawberry fields that are on s pretty severe slopes. And a lot of vineyards are on slopes and sometimes

winemakers actually love, you know, kind of the the challenged vineyard, you know, like that the this these Pinot Grapes had to struggle and therefore you can taste this or that. So there are a lot of different conditions that introduce a lot of variety and a lot of complexity. Sometimes you’re in a GPS-denied environment. Sometimes you have very low connectivity. So we see a lot of different things that come up. mud, muck, mire, all the above exist here. So this is like really a proving ground for physical AI. And there’s a feeling in ag in particular.

that if you can make it here, you can make it anywhere. And the vision is to have not just in Salinas, but also in like wine country and tell me more about that. So we opened Reservoir Farms in Salinas in March. And then a month later we opened our second location in Sonoma County. So Salinas is two hours south of San Francisco, but only about an hour south of San Jose. So it’s very accessible to NVIDIA to Google, etc. And they’ve all been here, which has been great.

And visited us. And then we’re in Sonoma County, which is an hour north of San Francisco. So we have startups that are splitting their time between YC and Founders Inc. and Reservoir, which is pretty cool. And that makes sense, right? Because YC, last time I checked, didn’t have a farm. So it’s useful to them. So yeah, we opened in Wine Country. We have fifteen acres of Pinot Grapes there. And it’s a similar structure that a startup can get onto our farm and be testing in rows and vineyards. And our plan is to

Roll out farms to cover at the very least the top ten US specialty crops by farm gate value. So wine grapes, table grapes, almonds, leafy greens, brassicas, carrots, tomatoes, blueberries, strawberries, and maybe I forgot a few more onions than they’re onions. And so then a startup that is thinking about working across ag, which is fragmented by crop, which is part of the complexity of this market.

they can work with Reservoir and then in one year or what a year and a half, they can actually access lots of different crops and test in lots of different environments. So Bonsai Robotics, which is one of the startups here that our fund actually invested in, they in the same week tested at our wine country location on vineyards and then tested on strawberries here in Salinas using the same machine. It’s pretty fun. So you’ll cover the whole supermarket. Yeah, we like to yeah we’ll open in Washington State sometime in the next six months.

And that will be tree fruit and blueberries. And we love Washington because it’s really like the intellectual and industrial epicenter of tree fruit and blueberries. The Cosmic Crisp Apple, which is my favorite apple. The Cosmic Crisp apple is legally by rules around the intellectual property, only can be grown in Washington. So yeah, so it’s licensed by Washington State University, which has the patent, and they actually license it to Washington growers.

And they could only grow it in Washington, I think in Clint 2032. Wow. But it’s an amazing, amazing apple. So then if you think about where would you build a startup in your startup space or where would you build a technical community that was focused on tree fruit, you would of course be in Washington, which is 60% of the apples in the United States. And then blueberries actually like there is extraordinary efficiency for blueberries in Washington relative to pretty much every other state. They grow more more inexpensively.

And similarly to these other environments, there is a lot of automation gaps. We don’t have hand harvest. We don’t have automated harvest in tree fruit. We don’t have automated harvest in fresh blueberries. So there’s a lot of opportunity, a lot of like big like really meaningful and high value tech gaps in these areas. While we’re doing our preparation, we talked about the operational layer of Archtek, right? And so there’s so many things that

goes in like beyond building the machine, beyond the road model and actually getting machines on the ground and keep them working. So talk to me more about that and what you’ve observed and what the gaps have been. Yeah, I mean the ab tech CEO has a lot to think about for sure. I think it really attracts it’s a very high demand sort of job. We tend to see more senior CEOs here because of that, not exclusively, but because of it, it requires a lot of resilience, a lot of humility. I mean there are a couple things. One is that the most

Critical thing is that they’re delivering a machine that really must deliver ROI within two years, maybe three, ideally one. So they are dealing with a very discerning customer that knows their numbers and is really kind of commanding a pretty high bar. It is a much more sort of sophisticated and informed sort of customer than I think a lot of roboticists are gonna be accustomed to. And that’s exciting. I mean, I think it’s a big challenge, but it’s exciting. And if it and if you can make it work, like

Carbon Robotics, for example, their laser weeder, they’re now doing a hundred million ARR. It’s great. So that’s a company that is is starting to scale in their other examples. But an AgTec CEO needs to think about the logistics of their machine, how they get it from here to there. They have to think about the servicing of those machines. Are they going to be servicing in-house with their own team? So Carbon Robotics, for example, is built an in-house service team. Bonsai is beginning to build one. And then but Bonsai also has what’s called dealer relationships.

So there are ag dealers that sell, may sell John Deere Tractors, and then they also sell ag tech as well, sort of more emerging ag tech. So they have to think about those sorts of channel partnerships, both for sale and service. They of course have to think about manufacturing and a manufacturing supply chain. So a lot of companies will do their own manufacturing for like maybe the first two or three years and sort of manufacture out of a warehouse, or they’ll work with specialized manufacturing firms or actually domestic.

that they come in a contract with, like Andros Engineering, which I’m sure you’ve never heard of. But if you go to Andros, which is in Passa Robles, our friends at Andros, you will see them hand assembling half a dozen different extraordinary AgTech machines that are all software enabled, AI enabled, and they’re building them. And there’s another one, GK Machine, which is in Oregon, which is doing something similar. And so you have these very specific manufacturing pipelines.

And addition to all the supply chain considerations they have to think about. So it’s a very opsy job. And then they have to think about how are they marketing their products and how are they pricing their product are they doing? Is it lease based? Is it service based? Is it, you know, selling the machine, etc.? So it is demanding that way and I would say I have a lot of respect for the ad tech CEO. It’s it’s not for the faint of heart for sure. I think one of the things that really stood out to me is how we’re farm is

currently operating like very closely with the community college around here. And I think that also touch upon the shifts in the labor market that we like as more autonomy is going to the industry, like what does the future like architect labor market look like? And what do we need the skills to be? And how do we train up the next generation of people operating on the field? No, it’s a big one for sure. And and we’re not the only ones talking about it, but

We think about we are building a technical community for agriculture and we’re doing it in place. Like we’re doing it in rural regions of the United States. So rural Washington, rural Arizona, rural California, and then we’ll go beyond that, rural New York, rural Florida, et cetera. And it’s difficult to imagine convincing the San Francisco engineer to want to be in place here. Like their ties are there, they’re part of that ecosystem. They’ve gotten accustomed to creature comforts of being in San Francisco. So when you think about

Who is going to be very affinitized towards rural California? Well, it’s the people who are from there. And so we have a labor pool here. We have a we have a high tech labor pool here that we need to work with the community to to train up. So we have to think about what are the hiring needs of a Carbon or a Bonsai or a Verdant Robotics. We have to work with the community colleges that are in region. It’s also the California state schools. It’s a little bit with University of California as well.

And we have to have a really like ongoing, I’m talking like weekly, sometimes daily dialogue about the sort of opportunities that go between those firms as we transition ag jobs to ag tech jobs. It’s one thing to be servicing and implement on a tractor that’s dumb, that’s just like mechanized. It’s like you just let’s say it’s like a big, dumb tilling machine, no offense to a tilling machine. But then going into something that’s smart, that’s AI enabled, that might have, you know, twenty four GPUs on it or something like that. Like we’re talking a

another level of complexity. We have a machine out there that’s an automated harvester that’s pre-commercial that has eight robotic arms. How many cameras? I don’t know, but like maybe a few dozen cameras. So there we’re talking about like a different different level of service orientation. So what we did is that when we began to build Reservoir Farms, the concept for it, which was two years ago, one of the first conversations we had was with the local community college here, which is called Hartnell College. It’s been around for over a hundred years.

And they are basically a feeder into agribusinesses here. Those have been historically like pest control advisors or you know, agronomists, they might be folks who help you figure out what kind of chemistry you should be using on your farm or figuring out kind of how to problem solve around disease control or weed control. But now those same sort of pipelines of agricultural talent need to be thinking about AI enabled machines and robotics with a hundred different degrees of freedom.

So that’s a really different thing. So how do we do it? We basically set an MOU with them about how we work together. And then fast forward to today, we have eight interns with our startups here from Hartnell College. We have eight community college interns at a deep tech, high tech startup incubator, which is incredibly cool. And so we view it like that there’s a strategic role of inclusive innovation in physical AI. If you’re building physical, if you’re building AI for the real world and you’re doing it in place.

You have to be thinking really early on about your labor pool. And we encourage companies to think about it. But then we as the sort of convener, as the technical community, we, you know, have to think about it as well. So we’re learning a lot, but it’s a it’s a neat and new mode. Yeah. And it’s really about kind of like mixing the like chemistry, like learning about the

the pesticide side of things, but also like how it interacts with like the autonomous and the world model side of things to use. It’s real. It’s like community integration. Exactly. So in the Central Valley of California, which is about two hours east of here, we are launching a version of Reservoir there next year. And we decided to launch it actually on a community college campus. Wow. So there’s a school called Merced College, which is Merced is basically two hours south of Sacramento and then two hours north of Bakersfield.

So then this really substantial stretch of California’s inland, Eastern California, let’s say, middle to eastern California, you have several of the largest ag economies in the country. So Kern County and Fresno County are both eight billion dollar ag economies annually. They’re massive. And so we thought, you know, we should probably build a Reservoir there. This is tree nuts, table grapes, processed tomatoes, carrots, etc. It’s like a really, really big part of the agricultural industrial sort of state.

We thought a lot about what would be the ideal scenario for building Reservoir there. And we decided that doing it at a community college farm made a lot of sense because community college farms are multi-crop because they’re trying to train students. So on one farm you have tree nuts, table grapes, citrus, stone fruit, processed tomatoes, melons, peppers, etc. You would never have that in an industrial farm. So then a startup can be thinking, How am I building my machine? Not just for one crop and one task.

Each new crop a startup adds is more TAM for them. For sure. So it’s like you can think add a crop, add TAM, add a task, add TAM. So this particular farm in the Central Valley is perfect for us. And rather than building, let’s say, you know, ten miles from that farm, we worked out a public-private partnership with a college to build Reservoir Farms right there. So we’re doing some different things and working on it a lot. I’m curious in terms of these inters working at your farm and also with all these startups that

being debated by your community, like what are the things that they are learning that they cannot learn in a a community college inside their classrooms? Yeah, it’s a great question. I mean they’re they’re learning about the very specific intricacies of of machine maintenance and repair. They’re beginning to see sort of hands-on design decisions about machines. They’re also learning the equipment and sort of how it’s used in the real world. So we have site managers and site coordinators that are employed by Reservoir.

all of which are are ag engineers, mechanical engineers that that know the way around a shop. And so they’re getting that pretty extraordinary hands-on experience. But a lot of times the startups themselves are. So if you think about what are the characteristics of a startup founder in ag tech, they may be a software wonk, but they may not be a machinist. And so very often we are teaching the re startup founders how to drive a tractor, how to work a forklift. One of the unique things about Reservoir is that we have tractors here.

for them to build on and test on and add sensors to and do all the things that they would need to do in an ACTEC context. I think we’re the only startup incubator in the world that has access to John Deere tractors for as part of our perks. You know? So like you know at White Combinator gets, you know, what you know a million dollars in tokens from OkenAI, we get access to a tractor. So I think it’s pretty unique. Take me five years out, right? So where do you see worse of a farm at? Like how many farms would you have and what are some

New use cases that you’re excited about. Yeah. I mean, I think we’ll certainly have farms across the American West to cover the largest crops. We’ll have expanded nationally. I think we’ll have also established Reservoir Farms and sort of the major international hubs that have some characteristic chal similar characteristics of challenges. So basically every first world market has a labor shortage. If they are also growing their own specialty crops growing if they’re in agricultural market, like sort of exporter versus importer, they have significant challenges. So it makes sense to do Reservoir elsewhere.

And there’s a lot of the due tech ecosystems that have formed around those areas, like Australia and the Netherlands and Israel and other places. So it just makes a lot of sense for us to be there. We would hope to see several solutions that are plugging gaps and harvest automation. I would think by like 2030, 2032, that we’re automating more and more of harvest, that we have automation of berries and leafy greens and vegetables and brush. So that’s a big one. I we also want to see the scaled adoption of sort of physical tools to replace chemistry.

So that’s UV light for disease control, that is laser weeders, that’s electric weeders, and a bunch of other novel solutions that we’re expecting as well. So that area, which is basically using atoms instead of chemistry in order to control disease, control weeds, that’s good for the world. It’s also necessary because we’re gonna see increased regulation around the use of chemistry, and it’s gonna become increasingly expensive to do chemistry at scale. And so

That’s something that we’re tracking and you know you’ll have to come back hopefully sooner than I. So awesome. Yeah. Well, thank you so much, Danny. This is awesome. I appreciate it, Michelle. Thank you so much. Thank you.

This morning NVIDIA announced Jetson Orin Nano 2, the entry tier of its robotics compute line. Put simply, Jetson Orin Nano 2 is a distribution strategy: NVIDIA’s move to lock in the physical AI value chain from the beginning of a product’s lifecycle. We will discuss in detail later in this piece.

Before that, we will first go over an overview of NVIDIA’s three-computer architecture in physical AI, and how the company is positioning for the next wave of growth in robotics and distributed computing. While NVIDIA frames its physical AI stack as three computers, two further layers come into play when building a working robot. Below training sit the data operations layer, and actuation and motion control. NVIDIA partners on both instead of building them, and we explore why. The entry tier is also becoming a crowded place, so we close on competing offerings from Qualcomm, AMD and others.

Sections

• What is NVIDIA’s three-computer architecture?

• What is a Jetson, and what do the numbers on the spec sheet mean?

• What did NVIDIA announce on August 25, 2026, and what gap does it fill?

• How does Jetson Orin Nano 2 compare with Thor and the rest of the line?

• Which layers of the physical AI stack does NVIDIA not sell into?

• Why doesn’t NVIDIA make robot actuators?

• Who competes with NVIDIA at the edge?

• What this means for an investor

• Claim ledger

What is NVIDIA’s three-computer architecture?

Deepu Talla, NVIDIA’s Vice President of Robotics and Edge AI, framed it this way on the launch briefing:

“Physical AI ultimately is the largest opportunity in front of us. And we’ve been working on this problem for over a decade. And we are starting to see now real uptake because of improvements in model accuracy and intelligence.” [1]

AI is moving into the physical world: factories, warehouses, farms and city infrastructure, and each of those settings needs models that run on the machine, not in a data center. NVIDIA reports more than 3 million developers building on its robotics stack and more than 10,000 companies on Jetson, either shipping a product or developing one. [2]

NVIDIA divides physical AI development into three steps, and assigns a computer to each. Train, simulate, deploy. A builder trains a policy for the robot to perform a task, tests that policy in simulation, then deploys it onto the machine.

Why does a robot need more than one computer? A robot itself is not one computer. Policy inference, perception, sensor fusion, limb-local control and safety monitoring are different workloads with different latency budgets.

A beginner’s guide to NVIDIA’s physical AI stack

Here is what NVIDIA sells at each step.

Train: DGX. Integrated GPU systems that live in a data center or are rented from a cloud provider. Training a robot policy on large volumes of video and telemetry carries the heaviest compute demand of the three steps.

Simulate: OVX and RTX Pro. Graphics-heavy systems built for physics simulation and rendering at scale. This allows a policy to be tested across thousands of virtual attempts before touching hardware. This is super important as robot hardware is costly to replace and breaks easily.

Deploy: Jetson Thor and Jetson Orin Nano 2. The modules that go onto the robot itself. These have to be small, fast and power-efficient, because a watt spent on compute is a watt not spent on runtime.

Around the hardware, NVIDIA gives away a large body of software. Two model families run across all three computers:

Cosmos. Open-weight generative world foundation models. This also includes tokenizers that compress video into the units a model predicts over, and a pipeline that turns raw video into training-ready synthetic data. It sits alongside Omniverse on the simulation computer.

GR00T. A robotics foundation-model family and data pipeline. It is trained on DGX, validated through NVIDIA’s simulation suite, and executed on Jetson.

For simulation:

Omniverse. A development platform for 3D simulation and digital twins, built on OpenUSD as its scene format. Omniverse Kit is the SDK.

Isaac Sim. A robotics simulation application built on Omniverse Kit, adding robot models, sensor models and ROS 2 support.

Isaac Lab. A robot-learning framework on top of Isaac Sim, for reinforcement and imitation learning at scale. It replaced Isaac Gym and ORBIT.

For deployment:

Isaac ROS. In NVIDIA’s words, “a collection of NVIDIA CUDA-accelerated computing packages and AI models” for building robotics applications. [3] GPU-accelerated ROS 2 packages, in plain terms.

cuVSLAM. A CUDA-accelerated stereo visual-inertial SLAM library, renamed from ELBRUS, that estimates a robot’s position from stereo cameras and an optional IMU. [4]

What is a Jetson, and what do the numbers on the spec sheet mean?

Jetson is NVIDIA’s product line for the deploy step, the modules that go onto the robot’s body. It is the embedded line of NVIDIA’s physical AI chip offering, built for machines that move: robots, drones, cameras and industrial equipment.

This is confusing, but Jetson Nano and Jetson Orin Nano are different products. Jetson Nano launched in 2019 on the Maxwell architecture and was the entry board. Jetson Orin Nano launched in 2023 on Ampere. The line goes Jetson Nano, TX2 for drones and small robots, Xavier for autonomous machines, then Orin and Thor. [5] And Orin is a family, with Orin Nano at the entry, Orin NX in the middle and AGX Orin above it.

A cheatsheet for compute terminology

TOPS or TFLOPS, with the precision attached. Trillions of operations per second. Precision is the number of bits used to store each number the model works with: FP32 uses 32 bits, INT8 uses 8, INT4 uses 4. Fewer bits means a smaller memory footprint and faster arithmetic, the trade-off is accuracy. The same chip produces roughly double the TOPS at INT4 that it produces at INT8.

Dense or sparse. A sparse figure assumes the model has been pruned so the hardware can skip zeros. Vendors tend to quote whichever number is larger. It’s important to check which density they’re referring to.

Memory capacity and bandwidth, in gigabytes and gigabytes per second (GB/s). For language and vision-language models, memory bandwidth, not raw compute, sets the limit on speed. This is why a part can have impressive TOPS and still produce tokens slowly.

Power envelope, in watts for the whole module. For a robot, this is battery life, and it converts directly into runtime between charges.

Form factor and module compatibility, in mm. Whether a new module drops into the existing carrier board determines whether an upgrade is a firmware update or a hardware redesign.

Supported-until date, calendar year. How long the vendor commits to shipping the part determines the length of amortization years on a design, which sets the cost structure of the chip.

What did NVIDIA announce on August 25, 2026, and what gap does it fill?

NVIDIA announced Jetson Orin Nano 2, the entry tier of its offering, on August 25, 2026. The specifications are below, all marked preliminary and subject to change on NVIDIA’s own materials: [6]

Compute: 78 TOPS INT8, Ampere GPU
CPU: 8-core Arm Cortex-A78
Memory: 8 GB LPDDR5x, 120 GB/s
Power: 15 W to 40 W, entire module
Compatibility: full-stack NVIDIA software, form-factor compatible
Availability: 1H 2027
Price: not disclosed. No price appears in the press release or the briefing deck.

Ampere vs. Blackwell. Ampere is NVIDIA’s 2020 GPU architecture. Blackwell is the current architecture, released in 2024. Here are the full Jetson generations: Nano on Maxwell, Xavier on Volta, Orin on Ampere, Thor on Blackwell. [5]

Orin Nano 2 is the first new entry-tier design in 3 years. Talla confirmed Orin Nano 2 is new silicon, still on Ampere, with the GPU optimized further for inference. [7] This means NVIDIA paid for a new chip design on a 5 year old architecture, instead of bringing Blackwell down to the entry point.

Jetson Orin Nano Super, launched December 2024, used the same Ampere silicon as the original Jetson Orin Nano from 2023. A software update increased performance from 40 to 67 TOPS, and lowered the price from $499 to $249. [8] Orin Nano 2 is an actual new chip at that tier.

Before Orin Nano 2, NVIDIA’s compute offering was strong at the high performance end and stale at the entry end. Built on Blackwell, Thor put a data-center-class part on the robot at 2,070 FP4 TFLOPS, 128 GB and 40-130 W. [9] The entry tier stayed on Ampere, at 40 TOPS on the original Orin Nano and then 67 on the Super. Both were too slow for a VLM.

NVIDIA shared real time model performance for Orin Nano 2 against Thor. Against a 20 tokens-per-second line the company labels real-time interactivity, 5 of the 10 models tested clear it. Every model at 8B parameters or above falls short, at 10, 12 and 14 tokens per second. The 4B tier straddles the line, at 17, 18, 21 and 21. The 2B tier clears comfortably, at 34, 39 and 42. Cosmos Reason 2 8B, NVIDIA’s own robotics reasoning model, produces 14 and misses NVIDIA’s own threshold. [10]

How does Jetson Orin Nano 2 compare with Thor and the rest of the line?

Orin Nano 2 is the entry offering of NVIDIA’s edge compute Jetson family. It replaces Jetson Orin Nano Super and shares its form factor. That means a builder can move to it without redesigning the carrier board.

On NVIDIA’s own benchmarks, Thor delivers 3.8x to 6.1x the throughput of Orin Nano 2. Thor executes 31B and 35B models above the real-time line. Orin Nano 2 tops out around 4B. [11]

Qwen 3.5 4B: 18 to 68 tokens per second, 3.8x
Gemma 4 E2B: 34 to 133, 3.9x
Gemma 4 E4B: 17 to 95, 5.6x
Qwen 3.5 9B: 12 to 73, 6.1x

Thor is the brain. Orin Nano 2 is the part for a gripper, a sensor head, an AMR base, an inspection box beside a line. NVIDIA demonstrated two Reachy Mini robots operating from a single Orin Nano 2, each executing a small language model, speech recognition and vision at once. [12]

Which layers of the physical AI stack does NVIDIA not sell into?

While NVIDIA frames its physical AI stack as three computers, there are more than three layers involved in building a working robot. Below training, there’s the data operations layer, the collection, curation, labeling and fleet observability that feeds everything above it. And underneath the machine sits actuation and motion control, the motors, reducers, drivers and the control loops that move the machine.

NVIDIA partners on both layers. Deepu Talla said on the launch briefing:

“So NVIDIA, we build these three computers, we sell these three computers, we build the acceleration libraries and reference models and software on top of these three computers for data generation, for training, for testing, and lastly, for runtime deployment. And then we work with literally every robot company on the planet, whether they’re building a physical robot, whether they’re building a robot brain, whether they are in the field of actuation or sensing, whether they’re providing just data collection services, it doesn’t matter what layer they’re in the ecosystem, we partner with every layer of the ecosystem. NVIDIA does not build robots, but we enable every company building this technology.” [13]

Talla names both missing rows himself, actuation and data collection services, 40 seconds after saying which layers NVIDIA builds and sells. He shared that actuation is becoming an extremely important part of the solution and that NVIDIA partners with several actuation companies. However, he didn’t name any companies. [14]

This gives us a rule that holds across NVIDIA’s physical AI strategy, and we will call it the silicon pull test: NVIDIA occupies every layer that creates demand for its own silicon, and partners on every layer that does not.

In data operations, fleet observability and semantic search over robot logs do consume GPUs. Those GPUs sit in a data center. NVIDIA already sells them through DGX and cloud providers. NVIDIA keeps giving away models and simulation, as long as it seeds and expands demand for its silicon. If data curation moves onto the robot, it creates a new compute socket on the machine, then the layer becomes NVIDIA’s territory.

Why doesn’t NVIDIA make robot actuators?

An actuator converts electrical power into controlled motion at a joint. A robot-grade one packages a motor, a reducer, an encoder and a driver into a single unit, and a humanoid carries 25 to 50 of them depending on hand complexity. [15]

Three reasons why NVIDIA doesn’t make robot actuators today.

1. Actuators fail the silicon pull test. Actuators do not create demand for new GPUs.

2. Actuators and chips have structurally different margins. Manufacturing reducers requires high capex and high fixed costs. Harmonic Drive Systems makes the strain-wave reducers that go into precision robot joints. In its FY2026 annual report, gross margin came in at 30.4%, up from 26.7% the year before. [16] Nabtesco reported its Component Solutions segment, precision reduction gears, at 10.6% operating margin on ¥45.5bn of sales in H1 FY2026. [17] NVIDIA’s gross margins are in the 70%s because of its fabless business model. [18] These are structurally very different businesses.

3. Cost per unit for actuators does not come down with scale. NVIDIA’s cost per unit of compute decreases with every process and architecture generation. That improvement comes from economies of scale over a fixed cost. It lets NVIDIA cut price per unit of compute, widen the market and hold gross margin in the 70%s at the same time. A harmonic reducer is the opposite. Its costs are materials, grinding tolerance and fatigue life. These are all variable costs. Increasing volume does not change the cost basis. Silicon can achieve annual price reductions. Precision mechanics cannot. That’s why we see a player of HDS’s scale achieving 26.7% gross margin, and break-even operating profit.

That being said, actuation increasingly requires real-time motion control, and that control needs silicon. Which may be NVIDIA’s playground.

Who competes with NVIDIA at the edge?

The entry tier of edge computing has become a crowded space. Qualcomm is shipping now. AMD released a Thor-tier product in July with a decade-long supply commitment. Hailo and Horizon are selling into the tiers on either side of Orin Nano 2. The common theme is that competitors are putting real-time motion control silicon on the same board as the AI accelerator, while NVIDIA keeps that function on a separate part.

Read more

The crowd didn’t gather around the humanoids. Queues formed at exoskeletons, at massage and acupuncture applications, and at robot performances. RoboCT demonstrated a 2.2 kg consumer exoskeleton with stated capacity of 200 to 300 units a day. [2] AgiBot’s research institute ran a scan-to-rent exoskeleton counter that passed 200 rentals in its first half day. [3] The widest-reaching product on that floor was a massage robot. Hangzhou-based EasyFuture said its massage robot has close to 1,000 partner stores and institutions across beauty, traditional Chinese therapy and sports rehabilitation, and that it covers roughly 40 minutes of a 60-minute technician session. [4] The force-controlled joints under it are Rokae’s, and Rokae supplies the arm rather than operating the fleet.

How we conducted the floor survey. Our representative worked 19 booths across halls A, B and C with the same four questions at each: customer mix, unit volumes, lead times, and pricing at quantity. Ten booths answered at least one question, three declined all four, six were not reached, and seven gave a usable mix answer. Answers were recorded verbatim in Chinese. Where a supplier declined, that is reported as a refusal rather than as an absence.

In this piece:

• What China shipped in the first half of 2026, and what component sales imply

• State Grid’s RMB 6.8bn robot program, and the 6% of it that is humanoid

• Where robots are actually working, and why three of them are pharmacies

• The dexterous hand category has more entrants than end demand

• Harmonic reducers and joint modules: two suppliers, two kinds of gap

• Open datasets and teleoperation as a service

• Wang Xingxing’s deployment threshold

What China shipped in the first half of 2026, and what component sales imply

China shipped over 40,000 humanoid robots in the first half of 2026, or 97% of the global total, according to the China Electronics Society’s 2026 Humanoid Robot Industry Development Report, released by Zhang Feng at WRC on 20 August. [5] That is about 2.2 times IDC’s count of roughly 18,000 global humanoid shipments for all of 2025, of which over 85% went into performances, education, data collection and guided tour services. [6]

What is inside the 40,000 is not published. We were not able to find more details beyond the 40,000 numbers. The report is not online, no definition of “humanoid” accompanies the number, and there is no split between bipedal and wheeled or between education and industrial use. [5] IDC, on the other hand, published in its 2025 humanoid numbers, that over 85% into performances, education, data collection and guided tours.

Unitree founder Wang Xingxing said at WRC that industrial task efficiency runs at 30% to 50% of a human worker. [7]

Supplier mix answers given on a show floor cannot be converted into robot counts, and our own survey shows why. Laifual (03952.HK), the Zhejiang harmonic reducer maker that listed in Hong Kong on 30 June 2026 and carried a market capitalization near HK$6.7bn (US$855mn) in August, told our surveyor at its booth that it shipped 70,000 harmonic reducers last year. [8] [9] Its listing materials shared 2025 unit sales at 292,000, revenue at RMB 260.9m and its China market share at 21.4%, second behind Leaderdrive at 27.5%. [8] A 21.4% share implies a China robot harmonic reducer market near 1.36 million units.

Omdia put 2025 global humanoid shipments at about 16,000 units, with China accounting for roughly 12,800. [10] At a vendor-stated 14 harmonic reducers per humanoid, those robots absorb about 179,000 reducers, or roughly 13% of that 1.36 million market.

Both Leaderdrive and Laifual told us on the floor that humanoid customers are about 30% of their book. [9] Comparing Laifual’s own 30% against its own filed shipments, the math doesn’t check out. Thirty percent of 291,000 reducers is 87,300, which at 14 per robot is about 6,200 robots. Versus Omdia’s 12,800, that would make Laifual alone the harmonic supplier to roughly half of every humanoid built in China, while holding 21.4% of the market and sitting second behind a larger competitor. Both cannot be true.

There are 3 potential reasons for the discrepancy. 1) It’s likely that 30% is a share of customer accounts, and an account can be a design-in, a sample order or a program that stalled before volume. 2) Reducers per humanoid may be well above 14, which is a vendor marketing figure rather than a teardown. 3) A large share of shipped reducers went into inventory and development programs rather than into robots that shipped.

State Grid’s RMB 6.8bn robot program, and the 6% of it that is humanoid

State Grid has started supplier selection for a RMB 6.8bn (US$950mn) embodied-AI procurement. The program covers 8,500 devices, of which 500 are humanoid live-working robots. [11] That is about 6%. The tender specifies those 500 units for live-line work. [11] The buyer writing the specification is a utility, which is a different customer from the factories most humanoid roadmaps are built for.

PowerChina disclosed a strategic purchase of 5,000 embodied-AI robots. [11] The coverage named neither the robot type nor the supplier.

Galbot’s deployment gave the clearest price signal for humanoids amongst the whole WRC floor. Galbot was named first-place winner of a RMB 235.92m (US$33mn) embodied-intelligence equipment tender in Yibin, Sichuan, for 380 wheeled robots, 80 heavy-load robots, 10 robot dogs and 30 robot retail pods. [11] Five hundred units, an average of about RMB 472,000 (US$66,000) each, and not one humanoid among them. The buyer is a state-owned energy management company, and the award was published on 23 June, before WRC.

One Chinese trade publication called out order quality directly, in a week when almost all coverage was positive:

“A non-binding letter of intent can be called an order, a framework agreement executed across several years can be called an order... at least 80% of so-called commercialization orders are fake, and that is already a conservative judgment.” [12]

While the 80% is the author’s own judgment not an audited figure, it is a useful flag on a common reporting practice.

One claim from this show demonstrates how it happens. Chinese coverage credited a company called Longmeika with a welding deployment at a Korean shipyard. Longmeika is a Chongqing joint venture established in November 2025 between a local optoelectronics firm and the Korean robotics company Neuromeka. The shipyard work is Neuromeka’s own: twelve collaborative robots for block-assembly welding with HD Hyundai Samho, signed in June 2024, and HD Hyundai Samho’s own announcements name Neuromeka. [19] The Korean parent’s contract was reattributed to its year-old Chinese joint venture, and the reattribution then circulated as a Chinese deployment.

Where robots are actually working, and why three of them are pharmacies

40,000 robots shipped in first 6 months of 2026 , and over 85% of the 2025 base went into performances, education, data collection and guided tours. [5] [6] Where did the deployments go? And which came with a named customer and a named site? Below is every one we could find at WRC, graded by what stands behind it.

The common theme behind 9 deployments. Fixed scene, single task, known objects, and a working window with no untrained public present, usually overnight. The machine is not adapting to the world. The world was already arranged for the machine.

• Robot Era has its logistics sorting robots in batch deployment with China Post and SF Express across more than ten logistics centers in five provinces, and describes itself as in the batch delivery stage. [14]

• Ant Lingbo runs night-shift drug sorting at a Guoda Pharmacy store in Shanghai, taking orders through a receive-to-deliver loop alongside pharmacists, with the store requiring no space modification. [13] The interesting part is the architecture: one model, LingBot-VLA 2.0, driving three different vendors’ hardware, from Leju, Stardust and Ant’s own R-2. Ant calls the stage commercialization testing, names one Shanghai store and gives no store count. This launched at WAIC in July and reappeared at WRC.

• Galbot states over 100 smart pharmacies, retail pods at more than 200 locations across 45 cities, a deployment on CATL’s production line, and 80,000 accumulated hours of real-world data. [13]

• Star Era built a 1:1 reproduction of an autonomous forward warehouse with JD on its stand, and took over 100 real orders through it on the opening day with no human intervention. [15] Neither JD nor Star Era issued a release, and the reporting describes the show-floor unit rather than a warehouse in operation.

• EngineAI at Luxshare’s Suzhou plant. Traces back to EngineAI’s founder speaking to a Shenzhen district paper on 18 August, before the show opened. Likely due to EngineAI in IPO process and Luxshare as a newly listed entity (002475.SZ and dual listed in HK), no plant is specified and no unit count is given.[16]

• UBTech’s Cruzr Y1 at Fuao Wei Taike. [17]

• Leju’s Kuafu 5W running three to four months. The model is real: Kuavo 5-W appears in Leju’s IPO filing at 200 cm, 156 kg, 40 degrees of freedom and eight hours of endurance. [18] The three-to-four-month claim came from a staffer at the stand, quoted by one outlet on 21 August. Leju’s official channels featured a different and better-sourced deployment, at FAW Hongqi from April 2025, with continuous-operation stability stated above 90% across 47 acceptance items.

Pharmacy is the solved environment that three companies found independently. RobbyAnt, Galbot, and RealMan through its remote operation network all showed retail pharmacy deployments. [13] . [19] A pharmacy has fixed shelving, a bounded SKU list, no safety-critical human contact, and a night shift with an empty store. It meets all four conditions at once, which is rare in retail. Three companies arriving at the same scene independently is evidence about which environments qualify.

This gives a clue on what environments are next for deployment. Forward warehouses and inspection rounds appear twice each above, and both meet the same four conditions: fixed layout, one repeated task, a known object set, and a window when the floor is empty. Those are the scenes to watch, and they are among the same categories LightWheel used to label its open dataset, more on that in the data section below. [21]

Not one deployment on that list is a walking biped. The machines are wheeled, armed or quadruped.

The dexterous hand category has more entrants than end demand

Hall B, the components hall, held at least eight dexterous hand and tactile vendors within a few aisles of each other: Linkerbot, Xino, Inspire, OYMotion, Dahuan, Tashan, PaXini and EngineAI all took stands there, several within two booth numbers of one another. [24] ZWHAND showed a range from 6 to 21 DOF and claims full in-house development, with roughly a dozen models on one stand. [9] Inspire launched a 24-DOF tendon-and-linkage hybrid hand at the show. [22]

The count of fingers is a commercial decision, and one company’s answer is visible in its hardware. EngineAI’s T800, one of the two machines our surveyor recorded drawing crowds dense enough to block an aisle, carries a self-developed three-finger hand at 7 degrees of freedom with tactile slip feedback. [24] Not five fingers, on the most-watched humanoid at the show.

An exhibitor at a neighboring component stand put the reasoning plainly: in a factory, where the scene and the task are fixed, a two-finger or simpler gripper already does much of the work, and five fingers earn their place where tasks vary. [9] That is the solved environment arriving from the supply side. The conditions that make today’s deployments possible are the same conditions that remove the reason to pay for a five-fingered hand. While dexterous hands are challenging in an engineering sense from the control complexity and anthropomorphism, it seems that even the merchants selling these hands admit that these hands have yet to find the product market fit in a factory at today’s capability. [24]

Tactile sensing moved from fingertips to feet. PaXini launched foot-mounted multi-dimensional tactile sensing and a fourth-generation sensor line built on a self-developed 6D tactile chip. [22] Foot-mounted tactile improves locomotion stability; finger tactile improves manipulation.

Harmonic reducers and joint modules: two suppliers, two kinds of gap

Laifual understated its own shipments by 4.2x. At its booth on the Friday, Laifual’s representative told us it shipped 70,000 harmonic reducers in 2025. [9] Its Hong Kong listing materials put 2025 unit sales at 292,000, with revenue of RMB 260.9m, up 142.2%, and second place in China at 21.4% share. [8] The booth answer came about eight weeks after the company listed with audited numbers.

EYou shows the same gap from the other direction. EYou told us it delivered 95,000 joint sets in 2025 with roughly an 8:2 humanoid split. [9] Chinese coverage from the same week reports an order book above 1.5 million sets and a RMB 1bn (US$140mn) Wuxi base planned at 3 million joint modules a year. [23] Its own shareholder puts it as a 15-fold increase over full-year 2025 shipments. [23] The planned capacity is more than 30 times last year’s deliveries.

Humanoid is a minority of the book at every harmonic reducer supplier with real volume. Leaderdrive, which Gasgoo identifies as the 27.5%-share leader that Laifual’s prospectus lists only as “Company A”, [8] put humanoid customers at about 30% on both askings, two days apart. Laifual said 30% humanoid against 20% industrial, 20% collaborative robots and the balance machine tools. Tianlian said humanoid is 5% of 200,000 sets. [9] The humanoid-heavy names are the small ones. For the suppliers shipping at scale today, humanoid is the marginal customer and industrial automation is the business.

Open datasets and teleoperation as a service

Many attendees criticized the show on the grounds that most demonstrations run on teleoperation, and the criticism is accurate. An engineer quoted in Chinese coverage put it plainly: “A ten-minute demonstration takes the team countless rounds of tuning beforehand. Before the performance, even we thought it would fall over.” [12] More datasets are being released as open sourced, while teleop-as-a-service is growing as a product category.

• Noitom released HiPHI, a human motion and human-object interaction dataset, free to the global research community. [1] Company-stated.

• Beijing Humanoid released Pelican-Unify 1.0 and the Tiangong Omni open platform, on a machine it states at 39 kg. [1]

• LightWheel open-sourced EgoSuite-Open100K with Hugging Face the week before WRC: 100,000 hours of egocentric human data, over 15,000 tasks and over 15,000 distinct collection scenes, spanning 128 scene types across seven environment categories including retail, logistics, hospitality and industry. [21] The first batch is live on the Hugging Face Hub with the remainder rolling out in stages.

• RealMan is selling teleoperation as the product. Its GLN remote labor network ran a robot restocking and dispensing in a 24-hour pharmacy, one on substation inspection, one making mooncakes alongside a Daoxiangcun baker, and remote operation of its own Changzhou factory from the show floor in Beijing. [20]

Wang Xingxing’s deployment threshold

Read more

Flagship conferences come with product launches. Here are the highlights of the launches that came out this past week and what they mean for where the industry is headed.

Can a robotics model learn a new task quickly and generalize to new situations?

GEN-1.5

The team led by Pete Florence at Generalist AI showcased the one-shot learning capability of its new GEN-1.5 model.

Source: GEN-1.5 launch blog post

The headline was eye-catching: robot policy now does in-context learning, like LLMs do when you paste an example into a prompt. And rightfully so, a robot picks up a task after a 3-12 second demo of a task it’s never seen, with no training at all. The results:

• 59% (±10%) average success from a single demo, no gradient updates

• 66.5% on a held-out task after one gradient step on 1 minute of data.

• 83% (±9%) after 10 gradient steps on 5 minutes of data, roughly 50 demonstrations.

Source: Generalist AI, GEN-1.5 post, August 2026. Across 10 tasks.

Robot policies have historically needed hundreds to thousands of demonstrations per task, collected on the target embodiment, then a training run. With GEN-1.5, the amount of demonstrations drops by orders of magnitude.

Why it’s important to the physical AI industry

Deployment economics

Currently, deploying robots commercially requires weeks of engineering per new task, or per new site. This release suggests that a robot can now be adapted to a new task or new site with minimal engineering, and only a limited amount of training data. That significantly shortens the time for a robot to get up to speed in a new environment and on new tasks.

UMI data wins over teleop

Jim Fan called it a “nail in the coffin” for teleop data.

“I’ve been saying for a while that teleop will not last, and GEN-1.5 is driving the final nail in the coffin.” Jim Fan, NVIDIA

UMI refers to a human wearing the robot gripper to collect data directly. Teleop is a human controlling the robot via a skeletal device or VR headset. Generalist has previously shared they used UMI grippers to collect over 270,000 hours of data, growing 10,000 hours per week. Companies like Sunday Robotics have been using this approach as well. The UMI data teaches robots “physical intuition”. It turns out that capturing the subtle sleight of hand we humans perform constantly with objects, the micro-adjustments, the feel of a part snapping into place, is actually key for teaching a robot the intuition.

Orangewood Labs’ OWL Arms

Built from stamped and folded sheet metal, with 25,000 hours of operational life, OWL is manufactured for deployment. With the FCC ruling, they’re moving the supply chain to America. Orangewood is a YCombinator-backed startup and has been deploying robots in the US and India. The reason I find this worth highlighting is the team built this with industrial deployment in mind, designing for ease of use and long performance cycles. Unit economics of commercial robot deployment are dominated by integration labor.

What robotics products were announced at Actuate 2026

Here’s a list of major launches from the conference:

Generalist AI: GEN-1.5, a foundation model that picks up new dexterous tasks from one short demonstration and transfers simulation demos to real robots zero-shot

1X Technologies: A developer platform, teased by Tom Sanocki, shipping soon

Orangewood Labs: A new industrial arm series built from stamped and folded sheet metal, rated 25,000 hours, preorders opening

Veeda: Out of stealth the same week under Sanja Fidler, building world-model simulation for robot training, live demo on stage

Dulo: Sebastian Thrun revealed a new robotics startup at the close of his keynote, still in stealth and, in his words, “very small”

Foxglove: An agentic data platform for physical AI, semantic search over unlabeled multimodal data built on NVIDIA’s Cosmos embedding model

Where were the robots at Actuate 2026?

Most people I talked to described Actuate 2026 as plenty of data infrastructure and not many robots. I can see why: other robot conferences like ICRA have dancing robots that steal the show. The conference was focused on the infrastructure layer of robotics, which is a nascent layer for a young industry.

The data plays were everywhere. Lightwheel had a large booth showing egocentric data services. NVIDIA demoed the HALOS stack. There were at least two teleoperation stations: Foxglove’s newly launched Remote Access, and Trossen’s newly released Rivet platform, which was quoted to me for $40,000 at the booth.

I highly recommend trying out teleoperation stations at conferences. I came away with new respect for teleoperators. It is physically and mentally demanding work.

The same robot burns the grilled cheese because it has no memory of having already started the grill.

Is memory the gating factor to robots getting deployed to industrial and home use?

At the same time, Figure and AgiBot have done days-long livestreams of long-running package sorting. How did they do it, if memory was needed for long horizon tasks?

This post is a deep dive into what robot memory is, what role it plays in terms of deployment readiness of robots, in industrial and home settings, and what the frontier is.

What does “robot memory” actually mean?

Recently there have been at least two talks about robots and memory. Chelsea Finn’s YC AI Startup School talk, where memory comes up as one hypothesis among several and is set aside, and the robotics edition of YC Paper Club “Why Robotics Still Isn’t Solved”, which had a presentation by Marcel Torne. Torne is a Stanford PhD student in Finn’s lab and built the work during an internship at Physical Intelligence.

When I looked deeper into the problem space, robot memory turns out to mean at least 3 different problems:

1. What’s happening inside one episode of an attempt. Did I already wash this dish? How long has this been on the heat? This is what the memory papers do. Status as of August 2026: five months old, and not solved. The numbers are below.

2. Improvement across attempts. The cheese was not properly melted inside the grilled cheese yesterday, even though it looked ready on the outside. I should lower the heat today. Status: started earlier, reports bigger numbers, on narrower tasks. Not solved either.

3. Preference that persists per person. This person prefers the shirts to be folded this way, the pans live here. Status: papers exist, and we have not found a robot shown doing it. The strongest real-robot work we found (VAP, POSTECH) identifies which object belongs to a given user. It doesn’t adopt a user preference yet.

How good is robot memory in 2026?

Feeding a model its own video is expensive, so the memory systems compress instead. π0 is built on PaliGemma, a 3-billion-parameter vision-language model, and it conditions on the current observation alone, with no history at all. Adding history means paying for every frame. At PaliGemma’s 256 tokens an image, three cameras and ten seconds at 20 frames a second is over 150,000 tokens for a single decision.

RoboMME is the bake-off: 14 ways of adding memory, each bolted onto the same robot brain, all scored on the same 16 tasks. Holding the backbone fixed is the whole point, because it makes the designs comparable to each other. With no memory, the policy scored 17.93%. The best memory design scored 44.51%, and several scored below the no-memory baseline, which means those designs made the robot worse than having nothing. The tasks run in simulation on a tabletop arm, which is the forgiving setting, no perception noise and no calibration drift, so 44.51% is the generous reading.

Memory is not a switch you turn on. Most ways of adding it make the robot worse. A few help.

Physical Intelligence went at problem 2 first, learning across attempts rather than inside one. Its π*0.6 model, trained with a method called RECAP, shipped in November 2025. The robot practises a task, its failures get labelled, and the next version trains on them. On the hardest tasks, throughput more than doubled and failure rates roughly halved.

Those are experience results, not memory results. They often get quoted as one thing. When someone says robots doubled their throughput on laundry, that came out of RECAP learning across attempts, which is problem 2. Problem 1, remembering inside a single attempt, is separate work: MEM, four months later in March 2026. MEM claims tasks that span up to fifteen minutes, cleaning up a kitchen or preparing a grilled cheese sandwich. On Physical Intelligence’s own bar chart, MEM reaches roughly 90% task progress on grilled cheese against roughly 50% without memory. Task progress is partial credit, not a success rate.

MEM’s number and RoboMME’s 44.51% do not sit on the same axis. RoboMME does not evaluate MEM at all. 44.51% is the best of RoboMME’s own designs on RoboMME’s simulated tasks, scored as success rate. MEM’s figure is task progress on tasks its own authors chose.

Other results run much higher, and the benchmarks do not overlap, so they cannot be ranked against each other. NativeMEM, published July 2026, compresses each past frame to a single token using the policy’s own vision encoder, and reports success rates rising from 32.4% to 84.0% in simulation and up to 98.7% on real robots. MemoryVLA, a year older, reports 84.0% across twelve real-world tasks. Each ran on a task set its own authors picked. Neither has been run on RoboMME’s benchmark, and RoboMME’s fourteen designs have not been run on theirs. So a buyer today cannot tell which memory design survives contact with their task.

Even RECAP, the part that worked, still has a person inside it. The practice rollouts were automated. Experts (humans) are still needed to label the rewards, specifying what counts as success. Teleoperators (also humans) step in mid-task to help correct the robot during the rollouts.

This is the frontier: the robot learns from its own failures, and a human has to tell it that it failed.

Why do factory robots work without memory?

1. The environment holds the state

Figure ran 249,560 packages over 200 hours, across three robots on autonomous fleet rotation. AgiBot ran 64,828 tasks at a claimed 99.99% on Longcheer’s Nanchang line. Both figures are company-claimed, from company-run livestreams, and neither company has documented any memory carried across episodes.

It turns out the production line doesn’t need memory. Package 12,000 does not depend on package 11,999. The fixtures hold the part in a known position, the conveyor belt delivers each package at a known time.

The robots averaged 1,248 packages an hour, about 2.9 seconds each, against Figure’s stated three seconds for a human worker. 249,560 packages is a quarter of a million repetitions of a three-second task.

2. The state is visible, and doesn’t need memory

Call it the photograph test. If you photograph a scene and send the photo to someone who didn’t know what happened a second ago, and they can do the next step correctly, the robot doesn’t need memory for that task either.

For example, for a half-folded shirt, anyone can look at that and continue the next folding step. Laundry folding doesn’t need memory, even though the task takes 10 to 20 minutes per load.

Grilled cheese on a pan is different. A photo of the sandwich cannot say whether it’s been on the pan for 30 seconds or 4 minutes, or whether the cheese is melting inside. It needs memory, even though the task is shorter, around 4 minutes long.

How long the task takes is not the determining factor of whether memory is needed. It’s also not about the complexity of the task. It’s about whether the current state is readable from a plain camera frame.

Just because a task has a long horizon, doesn’t mean it needs memory.

Which robot tasks are ready to deploy today?

Currently, the robot does not learn what is worth remembering. A person specifies it in advance, for every task, before training. Change the task and someone writes a new list.

Figure and AgiBot have shown robots can perform tasks for days. The question is which tasks the robots can do reliably at the current capabilities and limitations, memory included.

Below is a matrix of tasks, split into home and industrial use cases, and their deployment readiness.

Workplace, deployable now: structured logistics and fixtured production lines. Parcel induction and sorting, tote handling, palletizing, kitting, machine tending. Figure’s package sorting is exactly this. The line does the remembering, so memory is not the blocker.

Workplace, blocked on problem 1: unfixtured assembly and inspection rounds. Unfixtured means the part is not clamped in a jig at a known position: wire harness routing, connector insertion where the cable hangs loose, fitting flexible trim or seals. Inspection rounds: walk a facility, check 200 points, know which ones you already did and which reading was abnormal three rooms back. This is the near-term commercial edge of the research as of August 2026.

Home, “standard tasks”: These pass the photograph test. These are the chores a stranger could walk in and finish. There’s a universal judge of whether it’s finished or not. Vacuuming, wiping a surface, loading a machine one item at a time.

Home, blocked, by different problems. Cooking is gated by problem 1 first and problem 2 second: elapsed time appears in no frame, which is the grilled cheese from the opening. Tidying is not gated by problem 1 at all, because the state of a room is visible in one look. It is gated by problem 3 and then problem 2, because where things belong is a household’s opinion, and the robot has to hold that opinion and improve on it.

What has to change before robots work in homes?

Two milestones will unlock robots capability significantly.

Milestone one: removing the human from deciding what counted as success and failure.

Currently, a person needs to define success and to intervene on failure for the robot. In a factory, once a workflow is defined as success, for example a tablet passes quality inspection, that definition is spread across tens of thousands of repetitions of the same workflow. AgiBot’s run was 64,828 tasks and 17,625 tablets. In a house, the task distribution is much wider. Human labeling and teleop intervention are needed more frequently at home than in a factory.

ENPIRE (NVIDIA, CMU, UC Berkeley, arXiv 2606.19980, June 2026) is the closest thing to a human-free loop we found. Coding agents run the improvement loop on real hardware: automatic scene reset, synthesized reward functions, parallel rollouts across 8 bimanual stations. It automates the reset of the scene, and the verification of the attempt. Humans are left with one-time work: set safety constraints, provide a few minutes of success and failure demonstrations so the agents can synthesize a verifier, and sign off on the reset procedure the agents build.

So why is this not already running on every robot? Because it gets more expensive per robot as you add robots.

The coding agents read the video and logs coming off every station and write their reasoning back. Token cost grows super-linearly with fleet size: the paper’s token use tracks the linear projection up to four agents and rises sharply at eight. The agents spend their hours reading logs, writing code and debugging rather than running robots.

Human labeling costs about the same per robot however many you run. The agent cost per robot climbs. The unlock will be to figure out how to scale agent efficiently to perform labeling in a fleet.

Milestone two: fully online learning.

With the human out of the reward loop, the second milestone becomes possible. RECAP runs in batches today: collect attempts, humans label, retrain, redeploy. The next improvement is the robot updates from its own attempts continuously while it works, with no labeling pass and no retrain cycle in between.

So is memory the blocker, or the reward signal?

Memory tells the robot what happened. Reward tells the robot what a good result looks like. A robot can replay the last four minutes perfectly and still not know the grilled cheese is meant to come off golden. A person has to specify that, once per task.

Read more

Unitree priced its IPO on August 6, the first time a profitable humanoid maker has carried a public price. It is the first real pricing event for embodied AI as an asset class.

On August 7, the company ran a 3-hour roadshow, where chairman, CEO and CTO Wang Xingxing, CFO Wang Feng, and board secretary Fu Fenghua answered more than 360 investor questions. Pricing was finalized at Rmb150.80 per share, giving Rmb61bn (US$8.6bn) market cap. Trading under ticker 688836 is expected to open between August 17 and 21.

In this article, I share 5 key points management disclosed, an annotated transcript in nine buckets, and the full translated Q&A transcript for paid subscribers. Fun side note: because of the full 3-hour translation, this is the longest Core Matter post to date, over 50,000 words, 189 pages long!

Unitree shipped more than 5,500 humanoids in 2025, excluding wheeled dual-arm robots, which the company claims as first globally. Blended gross margin is 60.13%, with humanoids at 63.18% and quadrupeds at 56.72%. Overseas sales were 43.65% of main business revenue, and one investor put the US at 13-19%, which management did not confirm. Humanoids contributed more than quadrupeds last year on both revenue and gross profit, at Rmb548.27m (US$81mn) of gross profit (54.40%) against quadruped’s Rmb395.69m (US$58.66mn, 39.26%).

At 92.9x 2025 ex-nonrecurring P/E, or 219x on this year’s run rate, Unitree positioned itself as “AGI via embodied intelligence”. Wang compared the present moment to the early home-computer stage. The early PC market was already selling to paying business customers, while over 70% of Unitree’s humanoid revenue today goes to academia and research institutions.

1. Industrial application was 9% of humanoid revenue

One investor put industrial application at 9% of humanoid revenue for the first three quarters of 2025, and three separate investors cited 70%+ humanoid revenue from universities and research institutes. Management did not dispute any of the figures. It said that category includes many secondary-development customers, referring to companies buying a platform to build products on.

Named customers are State Grid, China Southern Power Grid, PetroChina, Sinopec, Baowu Steel Group, Amazon and BASF. These purchases are for “validated application”, with no units, contract values or dates attached. The list came in an answer about industry applications, which across the session means inspection, fire and rescue and public services. Those are quadruped scenarios.

Core Matter take. Management’s own framing of the 9% industrial revenue is a matter of sequence: research buyers develop the application, then industrial buyers scale up in purchase later.

This is framed as not just a Unitree phenomena but industry wide. Management stated humanoid demand globally today comes from research, education and commercial consumer, and scaled industrial deployment “still needs time”. Questions about the renewal rates of research purchases were unanswered six times.

2. Q1 2026 net profit fell 52.55%

Revenue growth decelerated from 332.64% for full-year 2025 to 68.49% in 1Q2026. Excluding one-off items, net profit fell 52.55%. Management guided 1H 2026 revenue growth at 35.62%-45.41% and ex-items profits down 6.43% to 21.97%.

Management named five causes, three on the decelerating revenue and two on the declining net profit. Revenue: a larger revenue base, “industry enthusiasm gradually cooling” and intensifying competition. Declining profits: higher R&D on an expanded team, and brand promotion including 2026 CCTV Spring Festival Gala. They declined to name a turning point in 2026 or 2027, reading the same prepared paragraph to at least three separate questioners.

Core Matter take. 3 of the 5 stated causes sit on the demand side, and two of those are industry wide. “Industry enthusiasm gradually cooling” is a demand condition across the industry. I’d watch for profitability at other humanoid players that are set to go public, including AgiBot, EngineAI, Leju, to see whether oversupply and price competition are playing a role.

Unitree shared elsewhere in the session that it cut humanoid selling prices in 2025 to consolidate its industry position, which is also what took humanoid gross margin down to 63.18%.

3. The actuator stack and what vertically integrated really means at Unitree

An investor asked point-blank what share of harmonic reducers is self-made versus purchased. Wang answered without using the word “harmonic” once. Instead, he described a high-bandwidth force-controlled quasi-direct-drive (QDD) planetary rotary joint: high-torque-density motor, low-transmission-ratio planetary reducer, high-bandwidth closed-loop torque control.

Asked separately to describe the joint, Wang added a detail: a planetary reducer with a tooth profile optimized for legged impact loads, developed in-house, alongside an in-house permanent magnet synchronous motor.

The management team declined every harmonic supplier question (LeaderDrive, Tongchuan, Zhongdalide, brought up three times). Separately, when asked why reducers were absent from the R&D use-of-proceeds scope, Wang confirmed they sit inside the robot-body category.

Regarding their vertically integrated strategy, Unitree shared more color on what’s made in house vs. supplied. Design of the core components is held in-house, and so are the core algorithms, the motion control system and the models. Manufacturing is not in house. “For most PCBA surface mount, injection moulding and some machining, the company uses an outsourced processing model.” Lidar, cameras and dexterous hands are bought in from external suppliers in multiple models. This is different from previous findings that Unitree also produce their own dexterous hands, on top of sourcing from Inspire Robotics. What Unitree does in-house is design, incoming inspection, and final assembly of whole machines and core components.

Core Matter take: Most published humanoid BOM models put harmonic reducers among the largest single line items in the joint cost stack. If Unitree’s route is genuinely low-ratio planetary with high-bandwidth force control, the cost curve for the highest-volume humanoid maker in the world does not involve harmonic drive capacity at all. This also raises the question of whether other vertically integrated humanoid companies are following this strategy. This affects both the BOM and demand projections of related A-share suppliers.

In addition, management shared that “the marginal cost decline curve is less steep than expected”, naming sensors, actuators and control systems as the costs keeping humanoid prices elevated. The margin drivers is mostly revenue mix. The scale is a secondary driver, coming mostly from assembly cost. Component prices are coming down more slowly than expected.

4. Five models in 11 months, and DeepSeek partnership

Unitree gave an overview of their 5 models: UnifoLM-WMA-0 open-sourced September 2025, UnifoLM-VLA-0 January 2026, UnifoLM-X1-0 industrial-grade in early 2026, WVLA2.0 May 2026, UnifoLM-OminiA-0.3 July 2026.

One deployment is named in the 371 exchanges. UnifoLM-X1-0 completed pilot deployment testing in Unitree’s own factory in early 2026, autonomously completing tasks such as joint motor assembly. Asked how many models are commercially deployed, management gave no count. Asked separately about the gap to Figure AI, it said everyone, including their international peers, is at the test-deployment stage.

DeepSeek took 933,390 shares for roughly Rmb141m (US$20mn) in the strategic placement, locked for 36 months. Company described it as a Strategic Cooperation Memorandum, joint R&D on high-performance general robots. Five investors asked what the partnership has produced, and all five answers gave the same scripted answer with no project, timeline or spend attached.

Core Matter take. Management named two main blockers to deployment: embodied large model capability, and the durability of dexterous hands. Motion intelligence and structural components are described as already broken through and in volume production. So on the company’s own account, the hardware it sells is finished and the two unfinished pieces are the model and the hand.

5. FCC Covered List

On the FCC Covered List, Unitree’s six models hold certification and fall inside the advanced-robot-equipment definition: G1, H2, R1, Go2, B2, A2. That is the entire disclosed product line. The company did not answer questions about the FCC’s grandfathering approach, whether the company will abandon the US market, or whether they expect Europe to follow suit.

Core Matter take. We covered the Covered List mechanics in an earlier newsletter (link here). The grandfathering question remains open.

The best of candor in the session was their response to an investor’s question of whether Unitree is a toy company selling remote controlled robots. Wang compared teleoperation to the steering wheel and brake in an autonomous car. He said the remote exists as the final “physical safeguard against safety incidents caused by model errors or perception system deviations”.

What stood out to me is that the Spring Gala appearances and consumer price ladder they shared voluntarily, unprompted (Go2 at Rmb9,997, G1 at Rmb85,000, R1 Air at Rmb29,900) built the public attention and retail investors now bidding for the IPO, while most revenue came from labs and academia.

Unitree priced a long-duration call on embodied AGI, framed against the early home-computer stage. Beyond the narrative, I’m watching these metrics: H1 2026 gross margin when the semi-annual report lands, and 2027 industrial pilot conversion.

Annotated transcript and full translation below. Disclaimer: the transcript is taken from the official source on cnstock.com, translation is AI-assisted, and may contain error. Core Matter take throughout this piece (and all pieces in this newsletter) is not investment advice. Please reach out if you spot any errors. Any input is appreciated.

Annotated transcript: ten buckets

Source: Unitree SSE online investor roadshow, 2026-08-07 (issue announcement published 2026-08-07, online subscription 2026-08-10). roadshow.cnstock.com/ipo/688836

371 exchanges: 5 opening addresses, 363 investor Q&A pairs, 3 closing addresses. Each bucket below carries the verbatim Chinese, an English rendering, and the Core Matter annotation. Exchange numbers refer to the full appendix.

1. Valuation

Investor 182*****376 (#353):

宇树科技以扣非市盈率 219 倍，市销率 35 倍的发行价上市，如此高于同行和行业的估值的价格是基于什么考虑的？

“Unitree is listing at an issue price implying 219x ex-items P/E and 35x P/S. What is the thinking behind pricing so far above peers and the industry?”

Wang Xingxing (#308, the fullest of three identical renderings):

按照2025年度经审计的扣除非经常性损益后归属于母公司股东净利润59,075.28万元计算，公司本次发行价格对应的市盈率为92.92倍。同时，按照2025年度经审计的营业收入169,926.93万元计算，公司本次发行价格对应的市销率为32.30倍。相比而言，行业内主营业务与公司相近的可比上市公司及拟上市公司多数尚未实现盈利，公司则实现了较强的盈利能力，尤其在人形机器人领域。

“Based on audited 2025 net profit attributable to the parent after deducting non-recurring items of Rmb590.75m, the issue price implies a P/E of 92.92x. Based on audited 2025 revenue of Rmb1,699.27m, it implies a P/S of 32.30x. By comparison, most comparable listed and pre-listing companies with similar main businesses are not yet profitable, whereas the company has achieved strong profitability, particularly in humanoid robots.”

Core Matter annotation. The gap between the two numbers closes arithmetically, and neither party in the exchange closes it.

Both of management’s figures reconcile exactly on pre-issue share capital of 364,017,906 shares. 364,017,906 x Rmb150.80 = Rmb54,894m; divided by Rmb590.75m gives 92.92x; divided by Rmb1,699.27m gives 32.30x. On post-issue capital of 404,464,340 shares the same inputs give 103.25x and 35.89x. The 35.89x is where the investors’ “35x P/S” comes from, so the questioners are on a post-issue basis and the company is quoting pre-issue. The two bases differ by 11.1%.

Neither figure is 219x. Post-issue market capitalization of Rmb60,993m at 219x implies net profit of Rmb278.5m, which is within 1% of 2025 ex-items profit reduced by the 52.55% Q1 decline cited at #018 and #236 (Rmb280.3m). The 219x circulating through the session is therefore a forward or annualised figure built off the Q1 2026 decline. Publishable form: the company answers a 219x question with a 92.92x number, and the gap is the difference between last year’s audited profit on pre-issue capital and this year’s run rate on post-issue capital.

Two further points from the same bucket. The defense offered is that the peer set is mostly loss-making, so the multiple is defended by the absence of a comparable multiple. And at #236 an investor put the Q1 numbers directly against the multiple, asking how 332.64% growth decelerating to 68.49% and ex-items profit down 52.55% supports 219x, and received the total addressable market answer with no earnings path and no comparable.

How the price was actually set (#230, Gao Ruoyang, CITIC). The one mechanical pricing answer of the session: preliminary bookbuilding by qualified offline institutions, the highest-priced tranche excluded, then bids, comparable company valuations, and “the lower of the four values.” What is absent from that list is any issuer forecast, because none was given anywhere in the session (#199, #209, #241, #246). The multiple was set against a bid book and comparables with no issuer guidance underneath it.

Expense context (#063). Administrative expense went Rmb13.33m to Rmb25.01m to Rmb399.60m, reaching 23.52% of revenue in 2025. Management names the components in order: share-based payment first, then employee compensation, then intermediary fees. A larger post-listing incentive plan is promised at #201 and #349, on the same line.

2. The 2026 earnings decline

Investor 186*****814 (#322):

上会稿显示，宇树科技的营业收入、扣非净利润增速已较往期明显放缓，预计上半年扣非净利润同比下降6.43%—21.97%，请问：当前增速放缓是短期周期性调整，还是行业生命周期进入成熟期的长期拐点？

“The filing shows revenue and ex-items profit growth slowing markedly, with first-half ex-items profit expected to fall 6.43% to 21.97%. Is the slowdown a short-term cyclical adjustment, or a long-term inflection into industry maturity?”

Wang Xingxing:

随着公司营收基数已大幅提升、行业热度逐步缓和及市场竞争日趋激烈，公司2026年收入同比增速有所放缓。此外……公司在机器人本体与结构研发、具身智能大模型、运动控制算法等领域持续加大技术研发投入与新产品开发，并持续扩充研发团队，带动当期研发费用增长；同时……公司借助2026年央视春晚等平台开展品牌推广，当期销售费用新增金额较大，使得公司2026年1-3月净利润较上年同期出现同比下降的情况。

“With the revenue base substantially higher, industry enthusiasm gradually cooling, and market competition intensifying, 2026 revenue growth has slowed. In addition, the company has increased technical R&D investment and new product development in robot body and structure R&D, embodied large models and motion control algorithms, and expanded the R&D team, driving R&D expense up in the period. At the same time, the company conducted brand promotion using platforms including the 2026 CCTV Spring Festival Gala, and the increase in selling expense in the period was large, so net profit for January to March 2026 fell year on year.”

Core Matter annotation. Five causes, three on demand and two on spending. Demand: a much higher revenue base, “industry enthusiasm gradually cooling,” and intensifying competition. Spending: R&D on an expanded team, and Spring Festival Gala brand promotion. Two of the three demand causes describe the industry and not the company.

The full disclosed 2026 path, assembled by the questioner at #334: revenue growth 332.64% for FY2025, 68.49% in Q1, 35.62% to 45.41% guided for H1; ex-items profit down 52.55% in Q1 and down 6.43% to 21.97% guided for H1. The cyclical-or-structural question was asked twice, at #322 and #334, and answered neither time. No 2026 or 2027 inflection was offered and no forward guidance of any kind was given in the session.

Three supporting figures. Selling expense went Rmb37.72m, Rmb59.16m, Rmb141.20m across 2023 to 2025 (#262), which puts a number on the Gala explanation; 2025 selling expense is 8.4% of main business revenue, and the company spends nearly three times as much on administration as on selling. Humanoid gross margin fell to 63.18% in 2025 partly on a price cut management describes as taken “to further consolidate its industry position” (#105). And at #091 management states the industry has “no pronounced cyclical characteristics,” a few questions away from the guidance attributing the slowdown to cooling industry heat.

The same prepared paragraph was read to at least three separate questioners. #367 shows the mechanism directly: a question addressed to Wang Xingxing by name was answered by CITIC’s Chen Xiying using the exact text Wang had given twice earlier.

3. FCC and the Covered List

Investor 186*****814 (#318):

宇树境外收入占比超40%，美国市场占比约13%至19%。美国FCC已发布人形机器人禁令。请评估：若后续产品无法销往美国，对营收和利润的具体影响？

“Overseas revenue exceeds 40% and the US market is roughly 13% to 19%. The FCC has issued a humanoid robot ban. Please assess the specific impact on revenue and profit if subsequent products cannot be sold to the US.”

Wang Xingxing:

公司目前在售的主要型号产品包括人形机器人G1、H2、R1以及四足机器人Go2、B2、A2等均已取得FCC认证，并属于前述公告所指先进机器人设备，因此前述政策变更目前不会影响公司现有主要产品在美国市场的继续销售。

“The company’s main models currently on sale, including humanoid robots G1, H2, and R1 and quadruped robots Go2, B2, and A2, have all obtained FCC certification and fall within the advanced robot equipment referred to in the aforementioned notice. Therefore the policy change does not currently affect continued sale of the company’s existing main products in the US market.”

The question the paragraph does not answer (#267, 游客37835):

美国FCC新老划断的产品认证授权举措，对于公司影响有多大呢？公司是否有措施更新现有授权版的机器人，进而保持在美国市场的产品竞争力，还是说公司会放弃美国市场呢？

“How large is the impact of the FCC’s grandfathering approach to product certification authorization? Does the company have measures to update its currently authorized robots and keep product competitiveness in the US market, or will it abandon the US market?”

Answer: the same paragraph.

Core Matter annotation. Given repeatedly in identical wording by all three executives, and it is the most specific non-financial disclosure of the session. Two details are load-bearing. The six named models are the entire disclosed product line, and the qualifier is that the policy change does not “at present“ (目前不会影响) affect continued US sale. The paragraph says existing certified models can continue to be sold. Grandfathering is a question about the models that come after them, and #267 addresses neither the update path nor the abandon-the-market alternative.

2025 overseas revenue was Rmb731.66m, 43.65% of main business revenue of roughly Rmb1,676.2m (#328). Quote the percentage or the absolute figure; the percentage does not resolve against the Rmb1,699.27m total. The 13% to 19% US share is the investor’s figure and management did not confirm it. Asked whether Europe might follow, no view was offered.

Two scripts for one risk (#332). Asked what adjustments the overseas strategy will make, a different investor drew a different answer: the 43.65% restated, the products described as having first-mover, full-category and cost-performance advantages, and a commitment to build technical service and sales points globally. No adjustment is named and no distinction is drawn between the US and any other overseas market. The company has two prepared responses to the same policy risk and deploys them interchangeably.

4. The actuator stack

Investor 游客79006 (#222):

想请教公司人形机器人所用谐波减速器，自研与外采的占比大概是多少，未来供应链会有怎样调整？

“I would like to ask about the harmonic reducers used in your humanoid robots. Roughly what is the split between self-developed and externally purchased, and how will the supply chain be adjusted going forward?”

Wang Xingxing:

公司主要采用高带宽力控准直驱行星回转关节路线……以中低减速比行星齿轮传动为基础，通过电机力矩高带宽闭环控制，使关节具备高响应、强背驱性与可塑阻抗的高动态力控能力，是一种”高扭矩密度电机+低传动比行星减速器+高带宽力控”的旋转驱动方案。

“The company primarily adopts the high-bandwidth force-controlled quasi-direct-drive planetary rotary joint route. It is built on medium-to-low reduction ratio planetary gear transmission, using high-bandwidth closed-loop control of motor torque, giving the joint high responsiveness, strong backdrivability, and tunable impedance for high-dynamic force control. It is a rotary drive solution of high-torque-density motor plus low-transmission-ratio planetary reducer plus high-bandwidth force control.”

The same architecture with no supplier question attached (#094, 游客69217):

公司自研高功率密度永磁同步电机以及针对腿足机器人冲击工况的齿形优化的行星减速器……采用全内走线+双路磁编码、IP68等级的密封，使得公司核心零部件可在 –20°C-55°C环境长期工作；该技术采用快拆锥面抱合结构，实现快速换装。

“Self-developed high-power-density permanent magnet synchronous motor, plus a planetary reducer with tooth profile optimized for the impact conditions of legged robots. Fully internal wiring plus dual magnetic encoders and IP68 sealing allow core components to operate long-term from -20°C to 55°C. A quick-release conical clamping structure enables rapid swap-out.”

Where the reducer is funded (#175, 游客83620, “是没有技术能力吗” / “is it that you lack the technical capability”):

公司募投项目中包括机器人本体研发项目，减速器作为机器人本体的关键部组件之一，也纳入了该募投项目的研发。

“The funded projects include the robot body R&D project, and the reducer, as one of the key components of the robot body, is included within that project’s R&D.”

Core Matter annotation. The question at #222 was about harmonic reducers and the answer never uses the word. It describes a low-ratio planetary architecture where torque control does work that gearing usually does. Three things make that hard to read as a deflection built for the harmonic question. #094 gives the same architecture when nobody is asking about suppliers. #175 places the reducer inside the Rmb1,109.74m robot body project, where the line item reads “high-power-density transmission systems” (#154), confirmed only because an investor asked whether the omission signalled a capability gap. And the sponsor’s own component list at #149 names reducers and dexterous hands as full-stack in-house development.

No BOM share is disclosed for any component anywhere in the session, from either direction. #252 asked what percentage of unit value the harmonic reducer represents and whether it comes from LeaderDrive or Tongchuan Precision. #299 asked for the electric drive’s share of whole-machine cost and received a specification with no percentage. #311 and #312 asked whether the motors are axial flux, twice, and drew a refusal formula reserved for motor topology; axial against radial flux is a first-order determinant of joint torque density and cost. #240 and #313 asked whether the shells and dexterous hands are aviation-grade aluminum, PEEK or carbon fibre composite, which is a materials question with no vendor in it, and were refused. #154 funds “multi-scale lightweight structural parts for legged robots” without naming a material.

The in-house boundary, in the company’s own words (#350):

公司机器人产品核心算法、运动控制系统、具身智能模型、关键核心部件设计坚持自主可控，部分标准化通用元器件实施市场化采购。

“Core algorithms, motion control systems, embodied intelligence models and the design of key core components are held to independent controllability, while some standardized general-purpose electronic components are procured on the market.”

And the manufacturing side (#219):

公司对于大部分PCBA贴片、注塑及部分机加工等工序采取外协加工模式，由外协厂商根据公司提供的原材料及相关技术要求进行外协加工。

“For most PCBA surface mount, injection moulding and some machining, the company uses an outsourced processing model, with subcontractors processing to the company’s supplied raw materials and technical requirements.”

What is held independent is the design of key core components. What Unitree performs in-house is design, incoming inspection of custom parts, and final assembly of whole machines and core components. Electronics assembly, plastics and part of the metalwork are subcontracted. Lidar, cameras and dexterous hands are procured externally in multiple models and supplied on to customers (#176, #219).

Two limits management sets on its own cost story. At #031, because no scaled supply chain exists and downstream applications are immature, “the marginal cost decline curve is less steep than expected.” At #086, development and production costs remain high “particularly for key components such as sensors, actuators and control systems, which keeps humanoid market prices relatively elevated.” At #198 the margin drivers are named in order, mix first and scale second, and the scale benefit described is falling unit product cost on the company’s own volume.

Why it moves the thesis. Most published humanoid BOM models put harmonic reducers among the largest single line items in the joint cost stack. If Unitree’s route is genuinely low-ratio planetary with high-bandwidth force control, the cost curve for the highest-volume humanoid maker in the world does not run through harmonic drive capacity at all. That reprices both the BOM and the A-share suppliers the market has attached to this name, and it raises the question of which other vertically integrated humanoid builders are on the same route.

5. The models, and what is deployed

Timeline as disclosed (#173, #300, #324). UnifoLM-WMA-0 open-sourced September 2025. UnifoLM-VLA-0 open-sourced January 2026. UnifoLM-X1-0, industrial grade, pilot-deployed in Unitree’s own factory early 2026. WVLA2.0 May 2026, fusing world model with vision-language-action. UnifoLM-OminiA-0.3 July 2026, a single model coordinating multiple task types in home and eldercare.

Substantiation given for the first two:

WMA-0 (游客13700):

该模型作为可交互的世界模型仿真器，可支持10-20轮的多步交互推演。

“As an interactive world-model simulator, the model supports 10 to 20 rounds of multi-step interactive rollout.”

VLA-0 (游客72844):

仅使用单一策略（One Policy）即可在真实物理环境中完成12个不同的操作任务。

“Using a single policy alone, it completes 12 different manipulation tasks in a real physical environment.”

The count question (#184, investor 187*****957):

这些模型目前有多少已实现商业化部署？

“How many of these models are in commercial deployment today?”

Wang Xingxing:

大家都处于测试部署阶段，公司的WMA模型与VLA模型目前同样处于持续的研发测试阶段，目前主要在自有工厂等试点场景进行研发测试、部署验证，具备了相应的技术储备。

“Everyone is at the test-deployment stage. The company’s WMA and VLA models are likewise in ongoing R&D testing, currently conducted mainly in pilot scenarios such as its own factory for R&D testing and deployment verification, with the corresponding technology reserve in place.”

The one named deployment (#324, investor 158*****581):

2026年初公司自研的工业级具身大模型UnifoLM-X1-0已在自有工厂中完成试点部署测试，可自主完成关节电机装配等任务。

“In early 2026 the self-developed industrial-grade embodied large model UnifoLM-X1-0 completed pilot deployment testing in the company’s own factory, able to autonomously complete tasks such as joint motor assembly.”

Core Matter annotation. Read #184 and #324 together. Asked for a count of commercially deployed models, no count is given. Asked about brain progress, one named model is described doing one named task on one site the company owns, and the qualifier survives in the Chinese: 试点部署测试, pilot deployment testing. The task is Unitree using its own robots to assemble its own joint modules. No customer, no third-party site and no named external deployment appears in 371 exchanges (#280).

The leveling claim is applied to a named rival at #284: asked about the gap to Figure AI on autonomous task execution, “the company and other industry participants at home and abroad are all at the test-deployment stage.” At #193 the two halves sit in consecutive sentences: brain-level embodied model technology is “in a fast-developing exploratory stage,” and the company’s overall capability is “already in the global industry-leading tier.”

No public milestone for the largest use of proceeds (#295):

具身智能属于前沿技术领域，研发进程存在不确定性。公司内部制定有对应的研发工作目标，若后续出现达到披露标准的重大研发进展，公司将严格按照监管规则及时公告。

“Embodied intelligence is a frontier technology field and the R&D process carries uncertainty. The company has set corresponding internal R&D work targets,” to be announced only if they reach the disclosure threshold. Asked at #273 when roughly Rmb2bn of model spend produces a commercialisation inflection, no date is given. Asked at #296 for milestone nodes and a commercialisation timetable on the Rmb2,022.46m project, the answer repeats the project description from #155.

Where the Rmb2,022.46m goes (#150, from the sponsor). Brain is cognitive decision, task planning and the embodied large model; cerebellum is motion control and whole-body dexterous motion; body is physical structure including drivers and communication bus. The concession attached: “weak compute support and a lack of high-quality data leave models insufficiently intelligent, which has become an important bottleneck on the industry’s development.” The money buys cloud-edge-device compute, large real-world datasets, a cost-effective teleoperation system, automated annotation, synthetic data, and a data engine plus model factory plus evaluation centre. The teleoperation rig is funded as a data-collection instrument. It reads against #177, where Wang’s argument for the humanoid form is a data argument: “robot AI models are data-driven and need large volumes of real human data, which is humanoid-shaped.”

DeepSeek (#202, #259, #270, #323, #348). Hangzhou DeepSeek AI Basic Technology Research took 933,390 shares for approximately Rmb141m in the strategic placement, roughly 11.5% of the Rmb1.22bn strategic tranche, locked 36 months (#323). The relationship is a signed Strategic Cooperation Memorandum covering three areas: cooperative R&D toward AGI, deep cooperation on high-performance general robots, deep cooperation on AI large models. Asked five times, the disclosed content never grows past those three headings. #348 is the best-drafted of the five, offering three concrete tests (a joint project, model adaptation, a prototype), a binary on the technical route (plug DeepSeek’s model into Unitree robots, or jointly train an embodied-specific model), and a request for a date on the first verifiable deliverable. It adds one clause, “jointly conduct cooperative R&D and product development,” and answers none of the three tests, neither route, and no date. #237 asked whether the model is fine-tuned from a general large model or trained from scratch and what that does to compute cost, which is the question the DeepSeek relationship would answer, and drew the standard exploratory-stage paragraph. Tencent is named once, at #274, as a strategic investor alongside DeepSeek.

The two blockers (#316).

面向工业与家庭场景大规模商业化应用尚待突破的关键技术主要包括”大脑”层面的具身大模型能力与”灵巧手”的精细耐用程度两方面难题，其中最主要的技术难题还是全球范围内具身大模型均处于早期发展阶段，泛化能力不足。

“The key technologies still to be broken through for large-scale commercial application in industrial and household scenarios are two: embodied large model capability at the brain level, and the fineness and durability of the dexterous hand. The principal problem remains that embodied large models globally are at an early development stage with insufficient generalization.”

Motion intelligence and structural components are declared already broken through and in volume production. On the company’s own account the hardware it sells is finished, and the two unfinished pieces are the model and the hand.

6. The toy jab

Investor 游客21844 (#302):

听讲贵公司是遥控玩具公司是吗

“I hear your company is a remote-control toy company, is that right?”

Wang Xingxing:

类似于现阶段的智能驾驶汽车虽已部署自动驾驶功能，但仍需要配有方向盘与制动踏板……遥控操作作为机器人的底层安全保障，可随时接管机器人的控制权……为防范机器人因模型错误或感知系统偏差导致安全事故提供最终物理保障。

“This is similar to how intelligent driving vehicles today have autonomous driving functions deployed but still need a steering wheel and brake pedal. Remote operation serves as the robot’s base-layer safety guarantee and can take over control of the robot at any time. In extremely complex terrain, when the communication link is abnormal, or when an unexpected logic error occurs, the operator can use the remote control, which carries the highest command priority, to rapidly and directly execute an emergency stop, forced hazard avoidance, or correction of low-level motion parameters, providing the final physical safeguard against safety incidents caused by model errors or perception system deviations.”

The command path, asked separately (#329):

通用机器人的行动指令可以来自于人工遥控操作与群控系统、导航系统，亦可生成于具身智能模型或大语言模型。

“Action commands for a general robot can come from human remote operation and swarm control systems, or navigation systems, and can also be generated by an embodied intelligence model or a large language model,” with the model-driven path described as “an important direction of development.”

Core Matter annotation. The most candid answer of the session, and it puts two things on the record that the rest of the roadshow works around: models produce errors, and perception systems deviate. The answer never states what fraction of demonstrated behavior is teleoperated.

#329 supports it from a different angle. Asked how a language instruction reaches the actuators, the ordering given is human remote operation and swarm control first, navigation second, embodied model or LLM third, and the model-driven path is future tense.

The same point arrives without the joke at #249, where an investor puts five questions in one and management engages almost none of them. Two land: the public perceives Unitree’s “entertainment attribute” as far exceeding its “technology attribute,” and many direct purchasers are university research institutions and performance and exhibition companies, customer types that struggle to form stable large-scale repeat purchase. The Gala point is answered by citing the Gala as a valuable platform. The repeat-purchase point is answered by restating that the short to medium term market is research, education and performance, which confirms the premise. The fifth question, whether Unitree becomes an arm maker like KUKA or a component supplier if real production-line deployment takes ten years or more, is not addressed, and the numbered structure is abandoned after point three.

The consumer price ladder was volunteered without prompting (#042, #342): Go2 at Rmb9,997 in 2023, G1 base at Rmb85,000, R1 Air at Rmb29,900. Management’s own framing of the Go2 price at #342 is that it was “the first time a robot dog was brought into the consumer electronics price range.” At #260 and #263 management steers away from consumer brand extension, saying technology and product are the core foundation of the brand and marketing-led initiatives will be judged prudently.

7. Customer mix

Investor 137*****488 (#358):

2025年前三季度，超70%的人形机器人收入来自科研教育领域，工业应用仅占9%。公司计划如何突破”实验室到工厂”的鸿沟？目前真实的工业场景复购订单有多少？

“In the first three quarters of 2025, over 70% of humanoid revenue came from research and education, and industrial application was only 9%. How does the company plan to cross the laboratory-to-factory gap? And how many genuine repeat orders are there from industrial scenarios today?”

Investor 137*****848 (#289):

公司人形机器人业务 73.6% 收入来源于高校及科研院所，以教学研发类采购为主。

“73.6% of the company’s humanoid robot revenue comes from universities and research institutes, mainly teaching and R&D procurement.”

Wang Xingxing, the demand answer given to both (#338):

在具身大模型技术水平与真实场景泛化能力逐步提升的过程中，全球人形机器人的真实市场需求目前主要来源于科研教育、商业消费等领域，距离工业场景的规模化部署应用仍需时间。具身智能通用机器人行业发展初期，科研教育领域对机器人进行的科技研发、应用开发、场景验证及教育培训，是通用机器人后续在商业消费、行业应用领域大规模应用的重要基础与前瞻需求。

“As embodied large model capability and real-scenario generalization improve, real global demand for humanoid robots comes mainly from research and education and commercial consumer, and scaled deployment in industrial scenarios still needs time. In the early stage of the industry, the technical R&D, application development, scenario validation and education work that the research and education sector does with robots is an important foundation and forward-looking demand for later large-scale application in commercial consumer and industry.”

The one substantive pushback (#333, investor 136*****517 asking how long to get below 50%):

报告中的科研教育是泛科研教育，包涵很多二次开发客户。

“The research and education category in the report is broad research and education, and includes many secondary-development customers.”

Core Matter annotation. Three investors independently characterised the same concentration: 9% industrial and over 70% research and education (#358), 73.6% from universities and research institutes (#289), over 70% (#333). Management challenged the magnitude in none of the three answers. The single challenge is definitional, and it belongs beside the number every time the number is used: the research bucket is “broad” and includes many secondary-development customers, meaning companies buying platforms to build products on. No split between the two groups is given and no timeline for getting below 50%.

The definitional point connects to the moat claim. At #357 the open secondary development ecosystem is named as a competitive barrier, and at #310, asked how Unitree holds off carmakers with consumer channels (Xiaomi, BYD, Li Auto, XPeng), the defence is “a systemic moat formed from a broad installed base accumulated over time, user habit, and ecosystem content.” The moat and the customer-concentration problem are the same fact seen from two sides.

The repeat-purchase question, six times. #249, #257, #283, #296, #338 and #358. At #283 the term is 续费情况, the renewal position, which is the single number that would settle whether the research and education base is an installed base or a run of one-time sales. Handle 182*****367 asked the research market four different ways (#257, #265, #281, #283) and was refused each time. At #296 management points to “the prospectus and the relevant content of the inquiry responses (反馈回复),” confirming the repurchase rate exists in the SSE inquiry response documents, a filing separate from the prospectus.

The only named customers (#305). State Grid, China Southern Power Grid, PetroChina, Sinopec, Baowu Steel Group, Amazon and BASF. Two qualifiers travel with the list. The wording is 验证应用, validated application, with no units, contract values or dates. And the sentence sits inside an answer about industry applications, which throughout the session means inspection, fire and rescue and public services, and those are quadruped scenarios. The question that drew the list was about humanoids being confined to performance and research.

The split stated directly (#307). Quadrupeds: power inspection, fire and rescue and emergency high-risk work commercialise “relatively faster in the short term.” Humanoids: the short to medium term is research, education and performance. That division runs through the whole session and this is the sentence that states it.

Correction on the earlier reading of #207. The three-way mix (”in 2025 the shares of research and education, commercial consumer, and industry application were close, each above 30%”) sits between a quadruped sentence and a humanoid sentence, and the Chinese supports either a company-wide or a quadruped-specific reading. The earlier version of this bucket built an asymmetry claim on the quadruped reading. That claim is withdrawn. What survives both readings: no humanoid-level three-way split is given anywhere in 371 exchanges, which is why every figure characterising the humanoid customer base came from an investor.

One wording tell at #282: the middle category is called 商业活动, commercial activity, where #207 called it 商业消费, commercial consumer. Commercial activity is the more accurate label for the performance and exhibition business.

8. Supply chain: the motor wall, and the consensus list

Three prepared refusal texts are in rotation: main suppliers and procurement amounts, main raw material procurement, and third-party cooperation. The appendix records eight verbatim instances of the supplier refusal and thirteen of the third-party cooperation refusal, plus a terser fourth formula and a fifth reserved for motor topology.

8a. The motor wall

The component Unitree most consistently claims as full-stack in-house drew seven questions from five investor handles, answered by three executives and one banker.

No fact about the motor supply relationship is confirmed or denied anywhere in 371 exchanges.

Three questions trace the shareholding structure behind it. #244 establishes that Wolong holds indirectly through the Jinshi Growth Fund. #287 asks which A-share listed companies hold Unitree through Jinshi and at what look-through percentage. #361 puts the question to CITIC’s Gao Ruoyang directly:

金石成长作为公司机构股东，背后有多家电机产业上市公司参与出资。除财务投资外，该类产业背景股东，会在供应链、工艺、零部件开发层面给公司带来哪些协同支持？

“Jinshi Growth is an institutional shareholder, and several listed motor-industry companies participated in funding it. Beyond financial investment, what supply chain, process and component development support do such industrially backed shareholders bring?”

Jinshi is CITIC Securities’ own private equity platform, so the sponsor is being asked about the industrial composition of a fund its own firm manages, and it gives the issuer’s third-party cooperation refusal. The question establishes without confirmation that multiple listed motor-industry companies are limited partners in a fund on Unitree’s register.

8b. The consensus supply chain

Named-company questions and their counts. This is the crowd-sourced list of A-share names the market has attached to Unitree.

Also refused: the Jinshi look-through (#287, #361), Henan supply chain participation (#245), the Chengdu municipal agreement and Jinjiang district facility (#248), local governments and industry colleges (#241), axial-flux motor use (#311, #312), enclosure and dexterous-hand materials (#240, #313), supplier concentration and substitutability (#253), and second sourcing on motors (#326).

Core Matter annotation. The questions carry information the answers do not. This is a consensus supply chain assembled from sell-side notes and stock-promotion channels, and it is the list of A-share names with Unitree exposure priced in. The uniformity of refusal is a disclosure policy and not evasion of any one relationship. Two observations sharpen it. #248 is a publicly announced government partnership with a named district and a named facility, refused on the same template as a component supplier question. And #326 asks a supply-security question against 190,000 units of planned capacity and receives a procurement policy, a qualified supplier admission system assessing lead time, quality, cost and service, without stating whether any key motor component has a qualified alternate.

Three things Unitree did volunteer cut against a simple full-stack reading. It buys multiple third-party models of lidar, cameras and dexterous hands to give customers configuration choice (#176, #219). It sells joint modules, dexterous hands, collaborative arms and perception sensors as a product category of its own (#233), so hands sit on both sides of the line. And for components where external process maturity is high and external cost is lower, it “will consider custom procurement” (#292), given in answer to a question about whether the tactile and force sensors inside the dexterous hand are in-house or bought.

At #370 Fu Fenghua closes the session by committing that disclosure will be “true, accurate and complete.”

9. Governance and control

Investor 135*****191 (#242):

公司设置了特别表决权机制，王兴兴先生在发行后仍控制约65.31%的表决权……公司打算如何向市场解释特别表决权与”保护中小股东利益”之间的关系？

“The company has a special voting rights mechanism, with Mr Wang Xingxing still controlling approximately 65.31% of votes after issuance. How does the company intend to explain to the market the relationship between special voting rights and protecting minority shareholder interests?”

Fu Fenghua:

公司设置特别表决权机制，是保障公司控制权持续稳定、防范二级市场恶意收购风险、降低后续增发稀释影响、确保长期发展战略顺利实施的重要制度安排……对于修改公司章程、合并分立、解散、变更特别表决权股份类别等重大事项，特别表决权股份与普通表决权股份实行同股同权。

“The special voting rights mechanism is an important institutional arrangement to keep control stable, guard against hostile takeover risk in the secondary market, reduce dilution from future issuance, and ensure the long-term strategy is implemented. For major matters including amending the articles, merger or division, dissolution, and changing the class of special voting shares, special voting shares and ordinary voting shares carry equal rights per share.”

The key-person question (#223):

请问您关于公司领导层面，您身兼ceo和cto，未来公司会有计划聘用专职cto吗？

“You hold both CEO and CTO roles. Does the company plan to hire a dedicated CTO in future?”

The answer describes talent systems and organisational mechanisms and commits to neither a yes nor a no.

Core Matter annotation. Wang Xingxing is chairman, general manager and CTO simultaneously, and controls approximately 65.31% of voting rights post-issue. 171 core employees participate in the strategic placement, with the No.2 asset management plan locked 36 months. DeepSeek took Rmb141m locked 36 months. The board is 9 directors, 3 independent, including 2 accounting professionals (#137, #138), and independent directors have raised no objection to any board resolution to date. The sponsor’s stated comfort on the team is that “core personnel are deeply bound to the company” (#214).

Three items sit alongside. An explicit dividend undertaking is volunteered at #084 that no questioner asked for, which is unusual pre-listing. A larger post-listing equity incentive plan is promised at #201 and #349, on the same line that took administrative expense to 23.52% of revenue. And at #161 an investor asked the sponsor directly to name the main risks found in diligence; the sponsor named none and pointed to the prospectus.

One prepared script across issuer and sponsor (#367). A question addressed to Wang Xingxing by name was answered by CITIC’s Chen Xiying using the exact answer Wang gave twice earlier at #178 and #179. At #359 and #364 the sponsor answers market-sizing and investment-value questions with the issuer’s own paragraphs verbatim.

10. The raise and what it buys

Investor 158*****793 (#250):

本次发行预计募资净额约59.17亿元，比原计划42.02亿元超募近19亿元。请问超募资金的后续具体用途是什么？是否会用于并购、产能扩建还是补充流动资金？公司对这笔资金的IRR预期是多少？

“Net proceeds are estimated at approximately Rmb5,917m against the original plan of Rmb4,202m, an over-raise of nearly Rmb1.9bn. What specifically will the excess be used for: acquisitions, capacity expansion, or working capital? And what IRR does the company expect on it?”

Wang Xingxing:

对于超募资金，公司将结合自身发展规划，在履行相关审议和公告程序后进行使用，相关信息请关注公司后续的公告文件。

“The company will use the over-raised proceeds in light of its own development plan, after completing the relevant review and announcement procedures. Please refer to the company’s subsequent announcements.”

Planned capacity (#232):

项目建设期 2 年，规划产能 19 万台/年。

“The project has a two-year construction period and planned capacity of 190,000 units per year.“

Core Matter annotation. The four funded projects total Rmb4,201.71m: models Rmb2,022.46m (#155), robot body Rmb1,109.74m (#154), manufacturing base Rmb624.11m, new products Rmb445.40m (#140). Gross proceeds at Rmb150.80 on 40,446,434 shares are Rmb6,099.3m, so net proceeds of Rmb5,917m imply issuance costs of roughly Rmb182m, about 3.0% of gross. The over-raise is Rmb1,715m net and Rmb1,898m gross, which is where the questioner’s “nearly Rmb1.9bn” comes from. Roughly 29% of the net raise has no stated use, and the IRR question is not answered. A separate question asking for the overall allocation of proceeds, which would have to account for the over-raise, points to the prospectus (#266).

190,000 units a year sets against 2025 humanoid shipments above 5,500 units and cumulative quadruped sales of 33,000 units across all of 2023 to 2025 (#036). Planned annual capacity is roughly 34x 2025 humanoid volume and about 5x the entire three-year quadruped history. The capacity-absorption half of the question, which is the half that matters at that number, is pushed to the prospectus.

Fixed assets are carried at Rmb35.14m against Rmb1,699m of revenue, 3.55% of non-current assets (#119), with right-of-use assets of Rmb57.86m (#129). “All current production sites are leased plants, with limited floor area and equipment configuration capability” (#152). Management’s own framing at #286 is the frankest line on the raise: “the company urgently needs to establish a scaled, intelligent manufacturing system.“ The manufacturing base is the cheapest of the four projects at 15% of the planned total and adds all of the physical output, while models take 48%.

At #118 an investor asked two things directly: why raise Rmb4.2bn against cash of Rmb1,419.26m (#113) plus Rmb282.23m of wealth management products (#114), and whether the shift in use of proceeds from buying buildings and sales sites to 85% R&D came from regulatory enquiry pressure or a genuine strategy change. The answer in full: “The fundraising is based principally on the company’s strategic needs. Thank you for your interest.”

Full translated appendix

Here’s the Full Chinese to English rendering of the 300+ Q&A from the official cnstock console. Primary source: roadshow.cnstock.com/ipo/688836.

Read more

Tactile sensing converts physical contact into data a robot can use. In GelSight's vision-based approach, a camera reads how a soft surface deforms and produces detailed information about contact geometry.

What is holding robot manipulation back?

In this conversation, Youssef Benmokhtar argues that the bottleneck is both hardware and data. The field has not converged on a standard sensing technology or shared model, which makes tactile data harder to combine at scale.

In this episode, I speak with Youssef Benmokhtar, CEO of GelSight. Tactile is on the edge of innovation for robotic dexterity. It’s the sensory bridge between computer vision and physical AI.

Youssef has extensive experience in imaging, optics, and scaling sensor businesses. He held leadership roles at semiconductor and photonics leaders like STMicro, OmniVision, and Magic Leap before taking the helm at GelSight in 2021.

GelSight, headquartered in Waltham, Massachusetts, is a spinout from MIT’s Computer Science and Artificial Intelligence Lab (CSAIL). The company also partnered with Meta FAIR to open-source Digit 360, a fingertip-shaped tactile sensor in 2024. In March 2026, GelSight was awarded a Phase II SBIR contract with the US Air Force to develop a miniaturized, rugged “digital fingertip” tactile sensor for intelligent robotic grasping.

As robotics move beyond locomotion to dexterity, high-resolution touch intelligence is increasingly critical for manipulation, maintenance, and hazardous industrial operations. GelSight describes itself as digitizing the sense of touch.

We talked through:

• The transition from vision to tactile intelligence: How digitized touch follows the same trajectory of visual optics and audio into trillion-dollar industries.

• Building a data-rich platform: Transitioning GelSight from a standalone hardware sensor to an integrated platform offering software libraries, cloud data analytics, and modular hardware architectures.

• Open source vs. proprietary software: The strategic rationale behind collaborating with Meta on the Digit sensor, seeding the academic research ecosystem, and standardizing tactile data for physical AI.

• The bottleneck in humanoid dexterity: Why manipulation needs an “ImageNet for Touch,” how platforms like NVIDIA Isaac enable Sim-to-Real translation, and how fine tactile resolution unlocks true superhuman precision.
  Navigating deep tech hardware strategy: Managing custom form factors, avoiding hardware commoditization, and setting clear ethical boundaries for robotics in defense and industrial automation.

Follow GelSight here: https://www.linkedin.com/company/gelsight/ | Website: https://www.gelsight.com

Subscribe now

You can also watch the show on Youtube

Every episode, I sit down with the founders, researchers, and operators building physical AI worldwide, tracing the full stack from components and supply chains to what actually deploys. See all published episodes here.

Chapters

• 00:00 - What is tactile sensing in robotics?

• 03:37 - How do you commercialize MIT research into industrial applications?

• 06:30 - What is a tactile intelligence platform?

• 08:46 - Is tactile data or sensor hardware the main competitive advantage?

• 13:56 - Why open source tactile sensors in robotics research?

• 19:07 - Why is tactile data the primary bottleneck for humanoid robot dexterity?

• 23:09 - How is the defense sector using tactile sensors for humanoid maintenance?

• 25:45 - How do deep tech hardware companies avoid commoditization?

• 30:41 - What are the ethical boundaries for robotics and military applications?

• 34:13 - When will tactile sensing reach a commercial tipping point in AI?

GelSight CEO Youssef Benmokhtar on why industrial surface metrology, not robotics, pays the bills; why open-sourcing DIGIT was Meta’s call and is now under review; and why he expects no single tactile foundation model.

You’ve literally visualized touch world has completely changed once vision was digitized. We focused on a vision of being the ubiquitous tactile intelligence company. I think we’re basically at the tipping point. We’re almost there. My guest today runs a company trying to digitalize the sense of touch. GelSight started as a camera-based sensor invented at MIT. Press it against the surface and it reads the geometry down to a single micron. fine enough to see a fingerprint. For years that lived in industrial inspection, checking aircraft skins and machined parts where a flaw isn’t an option. Youssef Benmokhtar took over as CEO in 2021. He’s not a roboticist. He came from imaging and optics Magic Leap OmniVision STMicroelectronics. A career spent turning sensors into businesses. His bet is that touch is the missing sense in robotic and GelSight’s job is to become the tactile intelligence layer, not just the sensor. Today, we’re going to talk about what it takes to give a machine a sense of touch, who pays for it, and where the value lands. Youssef, welcome to the show.

Welcome, Youssef. Super excited to have you on. Thank you, Michelle. We’re glad to be on. So you came from imaging and optics, Magic Leap, OmniVision, STMicroelectronics. What did you see in tactile sensing back in 2021?

You know, I always been in tech. I love tech. I’m curious by nature. Uh always loved innovation since my first uh role at STMicroelectronics. And when I got the privilege of meeting the uh GelSight co-founders, I had one of those other moments of, wow, this is so cool. Uh you’ve literally visualized touch and you digitize human touch. That realization, you know, just got me so excited about the potential of what you can do once you digitize human touch. I think the the decision to join the team and try to do something with the founders was u a natural decision. It was very fast. I was really attracted by the possibility to grow a business around digital touch. The potential for me is enormous because I had in mind based on my imaging background a really clear understanding of how the world has completely changed once vision was digitized. And you can say the same thing about when audio was when sounds right was digitized. It basically changed the way we live. It created a trillion dollar plus economies. It allowed for you know cameras to be embedded into phones into uh a bunch of things. You know we’re we’re talking through one of those digital cameras

right now. And when I met the team here, I thought maybe we can do the same thing with touch because really literally at the time you know before GelSight technology was invented there was no real digital touch out there. There were a lot of analog touch you know solutions but not a real digital touch and being vision based you know visual tactile sensor you know for me was more um you know even made it more attractive right because I I understand vision and I understand how the world has learned to leverage images and videos to create you know additional value through the analysis of those of that data and I thought you know having a digital based tactile sensor we’ll be able to leverage that existing baseline of talent out there that knows how to use computer vision and AI to analyze data. It was a really easy decision to join in 21.

Yeah, I can definitely see that connection of the dots there from vision to a vision based tactile sensor when you took over a CEO seat in 2021. In order to commercialize GelSight further from a lab technology to various applications, what changes did you need to make from how the company think about customers, what to focus on and what to build next?

Yeah, I think when I joined the company, what we had at the time was technology validation. We knew the technology worked. This was the work that was started at MIT and continued and refined at GelSight by the founders and a small team they had at the time. I knew I didn’t have to do anything about proving the technology works. What we definitely needed to do is where is the product market fit? Where is the technology can be applied today. There’s always been this this love story with robotics since the very beginning because obviously when you think about digitizing touch the first thing that comes to mind is well can I use that now to have make machines more intelligent and more capable you know handling objects and doing things like that. But the reality is digital touch has a lot of potential and the initial traction that we saw was more on the industrial market where our technology was a really good fit for uh surface metrology. So our focus initially was okay in order to pay the bills we have to grow this industrial market. We basically started to move towards confirming product market fit in industrial market. that has proven to be successful and now we’re more into that growth phase where you know we have a real product, real customers, growing demand and that’s that’s going really well. Now our our love for robotics has not gone away and I know we’re going to

talk about it at length today but we launched GelSight Mini and we collaborated with Meta on DIGIT exactly because we do believe that tactile sensing has a major role to play in robotics and especially for humanoid robotics and you know we had to find a way to be an actor in a very nascent industry. When you look at when GelSight Mini was launched, there were not all the big companies we talk about today in the human world that are raising, you know, insane amounts of money. We had to think about a strategy there in robotics as well, even if at the end of the day, what pays the bills more the industrial business. Yeah, for sure. You guys are definitely in a very unique position where you have that baseline covered with the industrial application and this is a wedge that you guys have always owned and then now there’s this new wave in terms of robotics and humanoid robots that are rapidly taking up these demand for tactile sensors. In 2022, you reframe GelSight as a platform to digitalize touch. So four years on, what does that look like as a business? Who are your customers today?

Yeah, we spent a lot of time really thinking about who we want to be. Many people in your audience are having the same kind of thoughts. We focused on a vision of being the ubiquitous tactile intelligence company. And what I mean by that is we know that referring back to what I said about digitization of vision and audio, the digital of touch is really something that’s going to allow us to be present in many many markets and not only the industrial market we’re in today. It includes robotics, it includes medical applications and includes future consumer applications.

So to do that, you have to start thinking about everything that you do needs to be aligned with that long-term vision of ubiquity. And you know why the tactile is because obviously we’re you know that’s our core expertise and we have to continue to be innovating on the pure sensing technology piece and the intelligence piece is where the platform came in because we ultimately do not want to be a sensor only company. We want to be able to offer value on top of our sensors and almost you know making the sensor not something our customers think about. what they think about is what is the value that you bring with your product you know are you solving my problem as a customer and that’s what we’ve been doing on the industrial market and that’s what we want to do in the future in robotics as well so the term platform was basically saying okay how do we basically build everything that we do starting with hardware which is not typically something you think about as a platform but by having a product that itself has this kind of modular aspect that it becomes a platform Chrome itself and this is where we launched the Modulus product last year because it is a product that is modular by design. It is something that evolves very quickly which is unusual for a hardware product and we do the same thing on the software side of things. We basically provide full applications that can run on top of the hardware. This is what we want to do in

the future in other industries like robotics. We want to take advantage of being kind of the first movers into the digital touch space that’s not own only the sensor piece but let’s also own the solution piece and when you think about the how to become ubiquitous where is the durable advantage do you see that is the optics the software or even the data side of things we have very very good understanding about how to make the sensing work uh we have very strong um IP around it but to your point I think At the end of the day, when it comes to the ubiquity, it’s going to be more about the data and what you do with it. The challenge is how do you provide as much data as you can but still be the one that know how to use the data the best. There’s kind of this weird balance that you have to find. If you look at, for example, what we’re doing today, whether it’s DIGIT or mini, it was purposely a strategy of let’s just make the hardware easily available and easy to use so that people can develop whatever they want on it. The idea here being that if you’re interested in tactile sensing, if you’re interested in dexterity, if you’re interested in object manipulation, we want people to think GelSight first. So, we had to make it very easy for that to happen, right? So, a small sensor, just USB cam, right? Plug it in, you’re ready

to go. You can develop your your code.

You can use Python libraries, for example, on DIGIT and GelSight Mini that were developed by Meta. But long term, we don’t want to stay there. We want to keep doing that but we also want to offer software solutions on top of our sensor including in robotics that would make to provide that solution like you know hey let me give you anti-slip solution that’s working best with the GelSight software versus something you might do open source let me give you libraries to do force estimation in the best possible way with GelSight you don’t have to invent it yourself use it and then build something on top of it as a customer that’s even better that’s where we’re heading in the future we have some things you know going on right now in um the development of our next generation sensor that are going to be basically prepping for that to happen where we’re not going to be only selling the sensor, we’re going to start sending libraries and other things to help people develop really valuable applications on top of the sensing. Yeah, that that’ll be really powerful because when it comes to tactile data is not just like visual or audio where tactile is like there’s so many data like there’s so many aspects of data that is streaming through like every single second. So making sense of it will be really helpful and to have various libraries to help people make sense of and how to make decisions based on these data will be very powerful and

works straight out of the box when you create that. So you brought a couple people on the board when you joined GelSight like Rony Abovitz from Magic Leap, Elliot Katzman from SolidWorks. What were you importing with those two and how did it change the company’s direction? Yeah. So, Rony, um, you know, I had the honor to work with him at Magic Leap. Actually, we’ve known each other for a while. What I really like about Rony, he’s a visionary. So, he’s someone that has this incredible ability to see the future and be able to articulate it in a way that makes sense to a lot of people. So, by bringing him on board, he really helped to define what the future of GelSight should be.

He was one of the first proponents I’ve ever heard use the term physical AI you know way before it became famous to these days and he was he kept telling me he said you are the necessary sensing technology physical AI how can we have physical AI without a sensor that is understanding the physicality of the world this was something that was really helpful in helping us to frame what we should be working on in the future and he’s been very instrumental in thinking about it in this way brought a different type of value for the company he’s somebody that you know every time I talked to him, he was sitting in front of me in this office and he was telling me Youssef, you’re not thinking about data enough. You’re not thinking about data enough. You have to think about data. And you know, initially, honestly, it was a very hard conversation because I said, but but Elliot Katzman, you know, the core of our technology is hardware. And he said, I understand. I don’t care. It should be ultimately about data. In his mind, he thought about Onshape and he thought about how you can reinvent CAD software thinking about how do you approach it you know CAD software through the cloud and shared a lot of his experiences at SolidWorks he was really instrumental in in helping us define the intelligence piece do not focus only on the sensing you know that’s great you’re already good at that keep you know stay ahead but how do you leverage the data that’s coming out of your sensors to bring that extra value

and that extra value cannot be local. It has to be something that’s available on the cloud. You know, something that a lot of people can take benefit from whether it’s within a company or within a community. So that’s probably what I would how I would summarize how Elliot Katzman helped to shape some of our roadmap. Yeah, that’s an really important piece in the strategy, right? where how do you evolve from being a hardware solution to I guess you alluded upon like having the data going to the cloud and then also like informing the decisions for the customers helping them work smarter helping the robots work smarter let’s talk about Meta DIGIT right so you’ve talked about open sourcing basic tactile intelligence walk me through the logic so what does open sourcing do for GelSight.

Yeah. So to be fair, the open sourcing strategy came more from Meta than from GelSight. We supported it because we were aligned in terms of of vision when DIGIT was launched in commercially distributed by us. Again, you’re talking about the infancy of tactile sensing and the use in robotics. And at the end of the day, if you believe in in the value that this technology brings that these products bring in the fields of robotic, you need to make it accessible and open sourcing made sense. This is what you want to encourage people in the community, these all these smart people that work in academia and corporate research groups to basically consider using these types of sensors in their work. Our vision has not been a commercial one when it comes to robotics. there has been more of a let’s seed the market and let’s make that seeding as easy as possible. The more people smart people use these sensors, the more they’re going to want to use them in the future and they going to want to base products on them. So that was the logic and the rationale behind the open source support I think from us and I think the strategy from the Meta DIGIT side. Yeah and definitely see that working in you know when you look at a lot of these research on tech sensing in the academic field most of these papers

use gels site or DIGIT in some ways so definitely see a lot of traction there to seed the markets when it comes to you know what stays open and what what stays patent and trade secrets where do you draw the line it’s a great question and I’m not sure Michelle I have the answer yet I think we’re basically at that tipping point, right, where the industry is, you know, initially, honestly, I’ve heard so many times, this is cool tech. I’m not sure we can use it even from these big companies that, you know, we all know these days. In a couple of years ago, two or three years ago, if you asked any of them, what do you think about adding touch to your hands, right, to your your robotic hands, most of them say, no, we don’t need that yet. And if you really look at the at the use cases was mainly pick and place you know warehousing type of things and so on. So we had a lot of push back to be honest you know people will say I want to play with it but I don’t have any plans look at it today almost all of the key players in the humanoid field are basically all about dexterity and digital touch and adding sensing capabilities to the hand. So look how quickly things change and I think the strategy we had on open source not only us I think other folks in academia as well is really helped to accelerate the integration of these kind of technologies and products but then to answer your question so that what does

it mean for us are we now back stuck into we’re only going to be potentially a sensing provider and not one that provides value on top of the sensor and I think this is the work that we’re doing right now prior to our future launch is to think about What else are we going to be offering on top of the sensor and what we still want to have offered as open source? I don’t think we have the answer yet to be honest, but we’re going to have to find some kind of a hybrid model. I don’t think that a fully closed system is going to be good for the industry or for us. But I also think that if we just leave it fully open, then we might be leaving value on the table for GelSight.

Yeah, for sure. And the industry is moving so quickly right now that kind of needs to evolve with their market. Right. So you mentioned a really interesting point where a couple years ago most robotics companies are not ready to use touch in their their work and with dexterity dexterous hands coming onto the market a lot like that actually open up the market for touch as well. So it seems that this is a very interesting we’re at this inflection point where there’s more and more demand for what you guys are doing. So the strategy needs to be adjusted real time as well. Yeah, absolutely. You know, three years ago at CES, I remember when you walked the floor in the robotic section, you had mainly arms or, you know, humanoids doing flips and other fancy, you know, things like that.

This year it was mainly about hands.

there of course there were more you know humanoid robots doing you know you know cool things but the number of companies ex you know showing hands it did not exist three four years ago and today’s it was all about that now I’m definitely biased maybe that’s what I’m interested in so that’s what I looked but those companies were not there three years ago they are absolutely were there this year and I suspect that next year we’ll see even more presence around this very important challenge of dexterity and object manipulation and and all those uh interesting topics. Things have changed for sure and we’re going to have to figure out our play our place also in that in the I want to talk a bit about data. So you touch upon that as well. So vision had ImageNet and touch has nothing like that yet. Is the bottleneck in manipulation now the hardware or the absence of tactile data at scale? I think it’s a bit of both because there’s definitely no model that seems to be the standard in tactile sensing. I think the industry is still looking for winner I would say for the sensing technology and then to develop the the right models from it. There are some things that have been published by uh by academia you know especially when it comes to simulation tools you know Taxim and things like that from Carnegie Mellon and MIT

has done some really interesting things. I think we’re getting there I think we are getting there. We’re not there yet. But you know just the fact for example that sim-to-real is possible on Isaac from NVIDIA you can if you want to simulate a GelSight sensing capability on your on your robotic hands and use their platform to do some sensorial work. I think when you have players like that really starting to deploy resources and offer solutions it just shows that we’re almost there. We’re almost there. It’s moving really really fast right now. But you’re right. There is no model today that I know of that would people would say that’s the that’s the equivalent of ImageNet for touch. But there are a lot of people working on it.

Yeah. And with touch there’s also this challenge where the tactile data is not standardized. Right. When we first t got in touch, we talked about standardizing the tactile data and that seemed to be also that bottleneck where how do you make sense of so much of these various data that’s coming from the tactile sensors, there’s optics, there’s magnetic sensors. So how do you go from the sim-to-real transition and correctly model the the behavior in real life?

Yeah. And and I think it’s also you know touch is a lot more complex than vision I think because even if you think about human touch capability right our fingers have a lot more sensitivity to touch than you know the the top of our hands or other parts of our body you know 100% of our body is a touch sensor but it doesn’t have the same you know resolution you know when we talk in terms of vision terms so I think it’s part of the challenge because some of these other technologies not vision based technologies like we have I think have a really good place in the in the robotics market. You know, whenever you’re going to need lower resolution touch, there are, you know, piezo approaches or other type of approaches that will be just perfectly fine. But at the end of the day, when you’re trying to build these models, right, you base it on one output at the end from all these different sensing technologies.

You know, if it’s GelSight, it’s going to be more of an image, so image based, so it’s going to be very rich, a lot of data. Others might be a lot lower, but it’s not the same type of data. So, I think the industry is still I think it’s trying to do everything as a single model and I’m not sure that’s going to work. I think it’s going to be probably, you know, you might need different types of models, you know, one that is really for maybe your generic touch type of application, low resolution, simple, fast, you know, that have different objectives. And then for really fine fine dexterity, super high resolution might be a very different model. But trying to do everything at once, I don’t know. Maybe one day we’ll get there. But to me today, it feels like, you know, we’re trying to um oversimplify the problem maybe.

Yeah. And do you see that GelSight is more occupying the the part of the market that is more high resolution touch for today? for sure just because of what we offer today as a vision based tactile sensor. It has so much information on every every touch that we’re definitely on the high resolution side. That makes sense today. Yeah. And so I want to talk a bit about your recent project with the US Air Force where you guys had this small business innovation research grants that is developing a compact tactile fingertip sensor. So tell us more about that and how do you see defense playing in the development of the markets in robotics.

Yeah. So so this grant is going to help us basically miniaturize you know our current GelSight Mini sensor. It’s also going to be you know something that’s going to help improve the overall performance of the sensor by a significant factor. The interest basically is how do you get as close as possible to mimicking human touch and it’s and human resolution you know in the air force and as well as many groups that do their own uh you know aircraft maintenance or you know rotorcraft maintenance a lot of the task are still done today by hand and I mean by hand is really feeling things uh feeling imperfections and and things of that nature. are other obviously solutions and and we offer it through our industrial products to to replace this kind of a lot of old way of doing things. However, you know the dream is if you now have humanoids one day doing the maintenance on an aircraft, wouldn’t it be nice if they can have the same or better than human resolution and then they can do things as they’re manipulating things around an aircraft.

So that’s the long-term vision is like how do I bring that kind of capability to have superhuman resolution in a tactile sensor. So that the human can not only use tactile to perform tasks, assemble, disassemble parts and so on, but also to characterize its texture, you know, to measure an imperfection, to, you know, don’t need a ruler or a caliper, just use your your finger and it tells you how wide something is and how deep something is. So that’s kind of the the long-term vision behind this development. So it’ll be a even smaller version of GelSight Mini and that that fits in the fingertips of a humanoid hand. That’s the hope.

Cool. And do you have like a timeline when it comes to that? How is that development going so far? It’s going well. You know, we are not at the liberty to give an exact timeline. The objective from the Air Force is really more of a more of a prototype objective, but we as a company are are have the ambition to take it not only to prototype but to actual production. We’re using the public funds really as a catalyst to get the program going, but our objective is to actually have commercial product at the end. So, it’s coming soon. We need a little more time. That’s uh exciting. So we’ve seen that tactile sensor cost collapsing in the past year and a lot of players are also vertically integrating into building hands themselves. So where do you see GelSight sitting when it comes to you know there are all these price competitors coming down from underneath and how do you avoid being commoditized?

Yeah. So you know I think you and I discussed uh before this challenge that we have in the industry that there’s no standard right. So you know when it comes to tactile sensing every humanoid company we’ve talked to had different requirements whether it’s mechanical requirements electrical requirements performance requirements and I think that makes it really really hard for independent you know uh sensor developers to be a one-size-fits-all kind of uh you know provider. So that’s why we chose the path with mini and I think even the one that we’re developing now we’re trying to make it so that it fits most of what we know but I also already know it’s not going to fit all because I’m sure some people have different are developing hands that will have other constraints that we that we’re not aware of. So I think there are two parts to your question. One is the question about commoditization and then the other one is how do you basically become also an actor with the people developing their own solution. So on the commoditization part, you know, I would say, you know, what one thing is you can’t stand still. I would be really worried if we didn’t have anything in the pipeline. We do have something in the pipeline. So I think as long as we offer that more value than what others might be coming up with, we should be good. We’re also designing it in a way that if we hit large volumes, the price

can easily go down. We also have you know very strong you know IP around at least the you know the visual tactile sensing technology that we have today. So there’s you know there’s kind of a defensive or offensive stance you can have if needed on that front you know compared to others that might be doing offering sensors in the space. But the most interesting thing honestly is really more how to work with these uh companies that have the interest and the means to develop these high-end hands that they’re interested in. And I think our technology has the advantage of being super flexible in terms of form factor and performance but still using the overall same concept from a technology perspective that I think it makes us ultimately I think a better fit for those that are looking for rich tactile sensor data in their hands.

Again, I’m not claiming that we’ll be the solution for all of their sensing needs, you know, touch sensing needs. But if you really want rich tactile sensing data output from sensors, you know, we are going to be the company that hopefully will be attractive for them to work with so that we can provide them with that solution instead of them having to develop it internally and kind of reinvent the wheel. We’ve been doing this for 15 years. Most of them, if not all of them, did not exist 15 years ago. We know what we’re doing. we have the right team, the right expertise to be able to bring value to these actors um in the robotics field. You know, we definitely will be looking for uh more, you know, partnership and collaboration opportunities with those companies just like we had with Meta a couple years ago.

Yeah, 15 years is a long time especially where a lot of these players are coming into the space like you know in the past year. So what one thing that you touch upon just now is there are so many different form factors and every hand might have a different requirement. So how often are you seeing when it comes to on boarding a new customers that GelSight team is adjusting the form factor or creating a bespoke solution to a specific customer. You know, so far we’ve been really trying hard to develop, I would say, a standard product, just to go back to your earlier question, a product that can satisfy the needs of most and we are using the data that people are willing to share with us openly to basically find the best solution for that. And that’s really where we’re focusing today. Now, we are having conversation with different companies that are saying, well, your standard product, even your future standard product might not be exact fit for what I need. So are you willing to have a conversation with us on a custom sensor and we are having those conversations as long as the business case makes sense for both parties right so you can imagine that there are some conversations but not I would say an enormous number of conversations because

the threshold to make sense you know business-wise is pretty high we are um still relatively nimble company so we cannot pursue all of the opportunities in front of us the threshold we’ve put in to engage on custom development is pretty high. Yeah. And I assume that these type of bespoke engagement would be asking for prototype units to start with which is super high cost and a lot of uh development resources from you guys as well on the technical side. So that would take up a lot of resources. You mentioned a lot about you know displacement and robots in law enforcements where you are against these type of things. Where did those convictions come from for you and how has it evolved over the years? This is something that was way before GelSight.

You know, we me personally uh I I’m hoping that the products that we develop are going to be helping humans for the good and that’s something that’s just part of my personal values and I’m trying to make sure the whole company is going to be operating under the same kind of values. So having robots that are going to be, you know, doing maintenance on aircrafts and things like that I don’t have a problem with. But having, you know, robots that, you know, say, “Hey, I need I need tactile sensing technology to be able to, you know, pull a trigger on the machine gun. H, no, I I don’t want to be part of of killing other humans.” It’s just something that we put as a as a moral bar, I would say.

And we’re trying to do that. It’s not easy. You don’t always know, by the way, what the the end application is. But my hope is that everything that we do is to make at the end of the day, you know, people on this planet have better lives, not not worse lives. So, I’m more about, you know, trying to find ways to augment people’s capabilities with these uh tools and and sensors. And if we really need to have a robot to do the job is because we can’t find the right skill set anymore for that job or you know or those jobs are not easy jobs and you know people should be doing things that are more fun and and those jobs we can leave to machines to do. That’s what’s driving us at least from a moral perspective or trying to.

Yeah, that’s super important. And if we think about various applications that robots can already better human lives like helping with elder care or remote surgery use cases that you mentioned and even like lifting heavy objects or working in dangerous conditions. These are a lot of applications that can very quickly contribute to like our well-being. And you also pointed out that it’s not easy for you guys to you know when you sell the tactile sensors it’s hard to really know what the humanoid robot end up doing and sometimes they might have shifts their strategies inhouse and they might go after things that you don’t know that they you may not be aligned to. So there’s I can see that there’s some internal challenges there when it comes to hey this is the belief and how do there’s also things that are out of your control.

Yes. Correct. Yeah. But you know we have customers with our industrial solutions in the nuclear energy right and when you basically take down a nuclear plant and you have to do maintenance you know time frame is very small it’s a very dangerous environment. Today it’s mainly people doing this. I mean there are starting to be some you know automated systems and vehicles and things like go do some things in those situations but these are typical cases where it would be a lot better for a machine to go do this you why would you want to send people in radioactive environments and dangerous environment where they have to time themselves right because they can’t stay more than x minutes in the environment these are the kind of things where if we had a robot that can go to pipe open a valve check something and close it right that’s something that I can a lot of value of a humanoid doing and if we can bring that kind of capability to allow those those tasks that would be great and then if you use our surface inspection metrology to even verify that you know how much corrosion you have and uh if there’s a a dent or a crack you can characterize that in the same time it’s even better definitely what is the moment that you’re waiting for that tells you tactile intelligence has arrived I think we’re almost there I think we’re almost there because Again I when you see the the number of papers that are

being published around this topic and when you look at the videos that are being produced the panels where the CTOs of these companies are talking about their next challenges that topic of tactile sensing you know or or dexterity or object manipulation it comes in different forms it’s just happening more and more and more so I think we’re almost there we’re almost at that tipping point where it’s becoming the challenge that the industry has to solve for the next step. I think it already has shown that it’s pretty good at, you know, if you want to move boxes around, if you want to do move parts around in a factory, it’s able to do that. A lot of different companies are offering those kind of of services these days. But if you want to do things that are a lot more complex, you’re going to need to add that dexterity. And I think everybody’s realizing it today. And I’m seeing a lot more interest, a lot more at trade shows, at conferences, number of papers published. So, we’re almost there. I don’t think we’re there yet. I think we’re probably still six to 12 months from being being there, but it’s soon.

Yeah, six to 12 from is uh very soon. And five years out, is GelSight the sensor company, the platform company, or a data company by then? I hope we’re also a data company in five years. That would be where we want to be. You know, the sensing company we already are will be continuing to develop and introduce new products. I I mentioned our Modulus platform. We we introduce a new hardware once or twice a year now, which is pretty nice cadence, you know, for hardware introduction. But really the holy grail for us is maximize the installed base and then start leveraging the data from the installed base. So five years from now definitely data amazing.

That’s our hope. Well, thank you so much. It’s great to have you on and uh I’m excited to see the upcoming new models coming into the market. Yes, abs I I definitely appreciate Michelle the invitation and the opportunity to chat with you today and uh yeah, stay tuned. It’s coming soon.

Most robotics builders and researchers we speak to put mass-produced home robotics years out. On the other hand, 1X says NEO ships in 2026, so does Sunday’s Memo. Weave says Isaac 1 ships in fall 2026. So, the next few months should put these machines into real homes, onto social media, and into mainstream news.

In this piece, we analyze the economic reality of these technologies, by working through these six questions:

1. What does the pricing actually signal?

2. How does each price compare to a human-equivalent hour?

3. How large is the subsidy, per unit and per trip?

4. What does an hour of real-home data cost, by channel?

5. What has to be true for these robots to deploy?

6. Who buys them, and how fast does the cost come down?

Pricing is the cost of admission to a home

In San Francisco, where I live, a three-bedroom cleaning is $200 to $250 per visit. Twice a month, that’s around $400 to $500 per month for a household line item.

That number may ring a bell. It’s what 1X’s NEO and Weave’s Isaac 1 both charge: $499 per month.

These two robots could not be more different underneath. Isaac 1 is a low-DoF gripper platform on a wheeled base with Physical Intelligence’s foundation model. NEO has 75 degrees of freedom, a tendon-driven bipedal body, a pair of top-of-the-line 25-DoF hands, built in house from body to brain. The bill of materials differs by a large multiple.

The price is not derived from cost, but from the household-services wallet share the company is trying to capture.

Where the robot lives

Nearly every home robot is teleoperated or tele-assisted. Where the robot lives determines the pricing strategy of the robot.

Resident: 1X’s NEO, Sunday’s Memo, Weave’s Isaac 1. Each robot belongs to one household. The customer purchases the robot. The company’s subsidy per household is capped by the price. Data cost per hour falls every month it stays.

Service: Tau, X Square. The robot arrives, works, leaves. Tau is teleoperated, X Square is human-accompanied. Every hour sold carries the same cost for the company, because the dominant input is a wage.

A resident deployment gets cheaper the longer it runs. A dispatched deployment does not get cheaper at all.

Rental. There is also the option to pay by month. Neither a NEO or an Isaac 1 subscriber ends up owning the robot. At $20,000 vs. $7999 for the same monthly rental rate of $499, 1X is either absorbing a much larger subsidy or expecting to need fewer human interventions.

Price vs. Cost per human-equivalent hour

In order to compare a $499/month with $30/hour with Rmb149/3 hours, we need to make some assumptions:

Utilization. Hours per month the robots actually work. At $499/month, 20 hours of use is $25 per hour. For reference, a home vacuum robot runs 45 minutes nightly logs about 22.5 hours per month, and it does one task with no hands.

Throughput. How fast the robot works versus a human. If a G1 takes an hour to do what a person does in 20 minutes, Tau’s $30/hour is a $90 human-equivalent hour, which is above the SF cleaner at $50 per person-hour ($200 across two cleaners for two hours).

Both dispatched products price below local labor. X Square at Rmb40.7 per hour, vs. Rmb80/hour for Shenzhen cleaner, Tau at $30 against $50 to $62.

The path to economic viability for dispatched service is by increasing the operator-to-robot ratio. Where staffing has been documented, the ratio is 1:1. AgiBot runs one operator per robot across 200 machines, a second operator is needed when a scene needs resetting. A Shandong data collection site staffs 33 collectors against 31 robots, a ratio of 1.06. Tesla pairs one robot with one motion-capture rig and one operator. All three are data-collection facilities rather than commercial services.

With an SF-based operator at $40/hour, at 1:2 the labor line falls to $20, which covers the $30/hour charge. Tau stated plans to move operators to cheaper countries.

How much is the subsidy

Sunday discloses that Memo costs $20,000 hand-built and targets 50% cost reduction at volume, implying sub-$10,000 price. The Founding Family program allows access to the units for free, so the subsidy starts at the full $20,000.

Memo is a wheeled, low-DoF gripper platform, similar to Weave’s Isaac 1. So its $7,999 is priced under cost at launch volumes.

NEO sits at the other end, with 75 degrees of freedom and two 25-DoF hands. Dexterous hands are the subsystem whose costs fall slowest with volume: a Sharpa hand sells at $50,000 each, $100,000 per pair (see our Dexterous Hand primer). A $20,000 price on that build is below cost on any reasonable reading. The question is how much.

Per trip, the dispatched version is about labor costs.

Tau charges $30 per hour. One operator per robot at $40 fully loaded, with idle time between jobs at roughly 1.5 operator-hours per billed hours, is $60. Add G1 depreciation of $11 per billed hour, on a $16,000 body over three years at 500 billed hours per year. Total is $71, less the $30 service fee, so about $40 subsidy by the company per billed hour.

X Square charges Rmb149 (about $22) per three-hour job. The accompanying worker minds the robot and handles the areas it cannot, so the wage sits somewhere between a robot technician and a cleaner. At ¥40 an hour loaded across 3.5 hours including transit, that is ¥140, which the ¥149 fee roughly covers. Platform depreciation is the rest: on a machine in the ¥150,000 range over three years at 500 billed hours a year, roughly ¥100 an hour, or ¥300 for the job. Net cost lands near ¥90 to ¥97 per billed hour, about $13.

So X Square is roughly at labor breakeven. If Tau moves its operators offshore, it’ll go to labor breakeven. But hardware amortization stays constant regardless of where the operator is. Accounting for that and maintenance, the hourly rate is still not breakeven in aggregate.

Sunday is the exception. It says that the robot collects no training data at all, and that Memo trains on recordings from its $200 gloves (more on that below). So, the Founding Family units and all sales buy no trajectories. They’re more a product-validation spend for distribution, proof of concept and failure discovery. That leaves Sunday with the most conventional economics in the group. Without the data trade to underwrite the discount, it has to reach a real gross margin on hardware eventually.

Cost per hour of real-home data, by channel

How much does the data cost for these home robots, and how does that compare with the data factories?

Resident. Assuming that the robot company is subsidizing each unit with $5000-$10000, a robot working an hour per day produces 365 hours per year. That puts raw data at $14-27 per hour. Not every minute yields a usable trajectory. At 60% yield, that becomes $23 to $46 per usable hour.

Dispatched. About $40 of net cost per billed hour for Tau, or $65 per usable hour with the same 60% yield. X Square is at $13 raw and $22 usable.

Facility benchmark. The comparison set is China’s purpose-built data factories, which exist to generate robot trajectories. Analysis published by Inside China’s Machine in July puts one usable trajectory-hour at ¥740 to ¥890 when a station runs at its best observed rate, and ¥3,990 to ¥4,800 at the utilization those facilities actually achieve across a year. The gap between the two figures is idle time: scenes being rebuilt, robots being repaired, stations sitting empty. At ¥7.2 to the dollar, that is $103 to $124 at best case and $554 to $667 in practice.

Counter-intuitively, the home channels are cheapest. X Square lands near $22 per usable hour, resident appliances at $23 to $46, Tau at $65, and a Chinese collection facility at $103 to $667.

A teleoperated home service generates data more cheaply per hour than a purpose-built collection facility, despite the facilities heavily subsidized. Every real home is a new scene, and each customer is paying something toward the operator wage. The facility pays its own operators, buys its own robots, and absorbs the utilization gap.

Sunday and 1X both are taking the robot out of the data loop. Sunday pays contractors, called Memory Developers, to wear a $200 sensor glove while they do their own chores at home. Sunday says it has 10 million of these recordings from more than 500 homes, and that Memo trains on those alone, with no teleoperation inside the customer home. Similarly, 1X released a glove kit with NEO’s tactile sensors and vision system.

What needs to be true for the home robots to work

Past the early adopters, these four conditions decide whether this crosses into the mainstream.

Safety. ISO 13482 for personal care robots, UL 3300, IEC 63310. A 35kg body moving near a toddler, elderly or pet either holds a certification or ships under a beta agreement and a liability waiver.

Privacy. Sunday claims that Memo “does not need to learn through human teleoperation in your home”. Weave discloses that the robot records visual content including people in frame, that its workforce may remotely access it, and puts the burden on the customer to configure the environment so it “only records visual content that you are comfortable sharing with Weave.” Tau retains indefinitely unless the customer asks for deletion. In this cohort, Sunday is the only company that does not ask the buyer to accept a camera with remote human access.

Maintenance and uptime. When it breaks, who performs the maintenance, at what cost. Mean time between failures, or a service-level commitment will move the unit economics significantly, beyond the price tag.

Functionality. How many tasks they can do reliably within the household.

Demand curve: iPhone or Google Glass?

Read more

Hi all, I had budgeted a slower newsletter week after last week’s Dexterous Hand deep dive (thank you for all the feedback, by the way!). With the FCC Covered List announcement, I thought it’d be interesting to dive into the components of these advanced robotics, and how the Buy American test might be reached. I’m not a policy expert, so feel free to reach out if I’ve missed anything.

Recap of what happened, ICYMI: On July 28, FCC released an updated covered list where any new foreign-produced advanced robots will be banned from importing to the United States.

“Foreign-produced” means failing the Buy American test. 48 CFR § 25.101(a): manufactured in the US and domestic components above 65% of total component cost, rising to 75% in 2029. It’s not retroactive, so models authorized before July 28 can still be imported and sold.

The rule is a domestic-content threshold on the bill of materials (BOM). This affects US robotics makers as directly as it binds Chinese exporters.

The key question is how many points of domestic content each subsystem moves per dollar of BOM, and whether a US-manufactured alternative exists at that layer today.

In this piece, I walk down the BOM one layer at a time, actuation, magnets, compute, battery cells, perception, radios and software, and explore what it means to builders and investors in robotics in the US.

Actuation decides whether the threshold is reachable at all

Actuators are 40%-60% of total BOM costs. Right now, most off-the-shelf actuators for robotics are made by Mainland Chinese manufacturers, eg, Daimon, CubeMars.

The Buy America test is where a part was made, not where the company is from.

Sourcing from Japan or Taiwan doesn’t help. The rule splits domestic from everything else, with no allied tier. Harmonic Drive Systems, Nabtesco and Shimpo fail identically to Leaderdrive.

Frameless and precision motors have “domestic” suppliers: maxon (through its 59,000-sq-ft manufacturing plant in Taunton, MA), Kollmorgen, Celera Motion, ThinGap. Roller screws also: Nook Industries, Creative Motion Control. Schaeffler has active manufacturing facilities in Cheraw, South Carolina for high precision components like rolling/needle bearings for linear actuators, and Danbury & Winsted in Connecticut for miniature ball bearings for micro-actuators in robotics.

Harmonic Drive also operates a 97,000 sq. ft. manufacturing facility in Beverly, Massachusetts, producing precision strain wave component sets, gear units, and rotary actuators. The question is whether their capacity will reach humanoid volume and price. A humanoid needs dozens of joints each at a fraction of that price, and the Beverly line is built for aerospace and semiconductor customers. There’s a gap in both volume and price for domestic capability to supply to humanoids.

Joint ventures are being set up by Chinese actuator companies to produce in the US. Minth Group (00425.HK) disclosed a framework agreement with Leaderdrive (绿的谐波, STAR 688017) on February 9, 2026 for a US JV designing and manufacturing humanoid joint module assemblies for North America. Their claimed target customers are Tesla, Figure AI, Boston Dynamics. That fits the Conditional Approval framing of “while they work to onshore manufacturing.”

Magnets

Sintered NdFeB is the input to every actuator on the list. China is above 90% of rare-earth refining and sintered permanent-magnet production. Below are some players in the magnets space based in the US.

Noveon claims to be the “only operational US sintered NdFeB manufacturer”. On the other hand, since January 2025, MP has been in commercial production of NdPr metal (the alloy input that gets processed into sintered NdFeB magnets) and trial production of magnets. Both claims can be true, since making the metal and making the magnet are different steps.

For scale, MP’s magnetics segment booked its first revenue in Q1 2025 at $5.1M and $19.9M in Q2, from precursor sales to GM under their 2022 agreement. MP also has the DoD price-floor arrangement.

Is magnet origin tested at the magnet or at the motor? There is no published guidance. That answer decides whether domestic magnets move a robot’s domestic share at all.

Compute

NVIDIA Jetson is the compute module most humanoid platforms we have priced are built around. A Jetson module is designed in the US, fabbed at TSMC, packaged in Asia. Under a manufacture-location test, this may not count as domestic.

Check country of origin on the module, not on the design.

Battery Cells

Batteries are a large BOM line. Robots run on small-format cylindrical cells, the 18650 and 21700 formats.

US lines producing those formats today:

• Panasonic Energy, Giga Nevada. 2170 cells at tens of GWh/yr, supplied exclusively to Tesla.

• American Lithium Energy, Carlsbad CA. 18650 nano-silicon, 4 Ah, coating its own electrodes in house, with a $13.2M California Energy Commission grant to expand.

• Forge Nano, Thornton CO. A 10 MWh/yr line running 18650 on a fully US-sourced bill of materials for DoD.

• Amprius with Nanotech Energy. SA128 21700, 6.8 Ah, 320 Wh/kg, built against FY26 NDAA requirements.

• Lyten, California. Lithium-sulfur cylindrical, dedicated defense and UAV capacity.

The constraint is captivity and price. Volume US cylindrical capacity is contracted to a single customer: Giga Nevada’s 2170 is the same format a robot uses, and all of it goes to Tesla. Merchant US supply runs at defense and drone volumes. Forge Nano’s 10 MWh a year works out to roughly 700,000 18650 cells, on the order of a few thousand humanoid packs. A robot maker can buy a US cell today at a price set by defense procurement.

Perception / Sensors

Perception sensor stack consists of depth, lidar and tactile, and the domestic capability varies by each component.

Depth has a US incumbent company and a foreign-manufactured part. RealSense, a spinout from Intel in July 2025 with a $50M Series A (Intel Capital, MediaTek Innovation Fund, led by a semiconductor PE firm), claims its depth cameras are embedded in 60% of the world’s AMRs and humanoid robots. Company-claimed, so treat it that way.

Intel’s regulatory page lists country of origin for every RealSense depth model. D405, D415, D435, D435f, D435i, D435if, D455, D455f and D456 are China/Thailand. The D555 is Thailand. That is filing-grade, from the manufacturer. Under a test that reads the production line, the depth incumbent counts as foreign, which puts depth among the more exposed layers.

Two other suppliers in the category: Orbbec (Shenzhen, the licensed successor path after Microsoft discontinued Azure Kinect) and Stereolabs (French, ZED cameras, acquired by Ouster).

Lidar is where the gap is widest. RoboSense (HKEX: 2498) ranked #1 globally in robotics lidar in 2025, over 303,000 robotics units and over 3,400 robotics clients. They shipped 185,500 robotics units in Q1 2026, up 1,458.8% YoY, about 56% of total shipments. Robotics passed automotive for the first time in May 2026.

Ouster (NASDAQ: OUST) assembles in San Francisco’s Mission District for new-product introduction and select US customer contracts, and runs volume production at Benchmark Electronics in Thailand, where it has manufactured since 2017. A US line exists at this layer and the volume does not come off it. Both companies’ figures are disclosed and the segment definitions differ, so read the comparison as order of magnitude rather than share.

Tactile: GelSight manufactures domestically. This rule is a direct demand event for them. The category problem is that no tactile benchmark exists, so buyers can’t compare parts on a published standard.

The radio module, and an exit from the FCC ban

The determination defines an “advanced robotic device” as needing all three of: sensor, connectivity component at 200 kbps or above, and software controlling navigation, perception, data collection or remote command and control. The 200 kbps is a criterion for inclusion.

Each of the volume commercial module suppliers below fails the domestic-component test under 25.101(a).

Chinese makers: Espressif (Shanghai, STAR-listed, ESP32 family), Quectel (Shanghai, cellular IoT modules, volume leader globally), Fibocom (Shenzhen, cellular modules).

Non-Chinese: Nordic Semiconductor (Trondheim, Norway), Murata (Kyoto, Japan), Telit (Italian-American).

US-manufactured radios exist, and drone makers already buy them. Silicon Labs is Austin-headquartered and fabless, so its parts are manufactured offshore. Silvus Technologies (Los Angeles, now Motorola Solutions) has the StreamCaster 4400 Enhanced on the DIU Blue UAS Framework, DoD-certified for military drone operations. Persistent Systems (New York) ships the MPU5 under the DoD Swappable Radio Module spec. Doodle Labs’ Mesh Rider is on the Blue UAS Cleared List and flies in Freefly’s Astro.

Those are defense mesh radios at defense prices, a different part from the Wi-Fi 6 and Bluetooth module a commercial robot carries. The nearest commercial analogue is Ezurio in Akron, Ohio, which designs in the US, and whose manufacturing location we have not confirmed. The silicon underneath any of these is Qualcomm, NXP or Infineon, fabbed and packaged in Asia. So it doesn’t count towards the domestic component cost defined by the FCC.

If a robot ships without a radio, or below 200 kbps, it falls outside the definition and isn’t blocked from authorization.

The OS layer, its precedent, and its ceiling

The drone ban set a precedent. AuterionOS is the cyber-secure software layer that let OEMs ship an NDAA-compliant, Blue-certified aircraft, with Auterion’s own Skynode and Skynav listed as Blue UAS Framework components. Auterion made compliant aircraft possible for OEMs building to the standard. It didn’t make a non-compliant imported airframe compliant by reflashing it.

Applied here, a US software stack is plausible evidence inside a DoW Conditional Approval application. It doesn’t move the domestic content percentage, because 25.101(a) counts articles and materials incorporated into the end product.

The Robotics OS layer is three layers, each doing different jobs:

• Onboard OS and middleware: ROS 2, dimensionalOS, and NVIDIA’s Isaac ROS. (Isaac Sim and Isaac Lab are simulation and training, not onboard software.)

• Fleet ops, teleop and data plane: Viam (founded by MongoDB co-founder Eliot Horowitz, NYC, describes itself as a container runtime for hardware, 200+ component drivers plus fleet management and data capture) and Formant.

• Observability and tooling: Foxglove, which is robotics data recording, visualization and debugging.

On July 29 (1 day after FCC Covered List news), Dimensional OS, an SF-based robotics software company, claimed they are now the compliant gateway for robots to access the US. Founder Stash Pomichter said there’s a different legal pathway from the Buy American test. As of mid-August the legal pathway referred to has not been published.

Security Certification Layer

No standard we found certifies an autonomous mobile platform as a whole. The adjacent ones each cover a neighbouring scope: IEC 62443 is industrial control, ETSI EN 303 645 is consumer IoT, UL 2900 is generic connected products.

A robot security standard, and an accreditation layer that tests against it, would add transparency and accountability to this nascent field. This layer can either be a standalone product, or a sub-category within the operating system layer.

Assembly is not the loophole

The Buy America test measures component cost. Labor sits outside it. Moving final assembly to the US does nothing to move the domestic share. This makes sense against the document’s stated goal in its first paragraph: reindustrialization of the US economy.

A lot of outstanding questions remain.

Does the 41 U.S.C. 1907 COTS waiver travel with the borrowed 25.101(a) definition? COTS means commercially available off-the-shelf, a commercial item sold in substantial quantities commercially and offered to the government unmodified. Most robot components are COTS. If it carries, the threshold is far easier to clear than it reads. If not, very few robots on earth qualify.

How fast will DoW process Conditional Approvals, and against what? The criteria aren’t published, so we’ll see more granularity as OEMs start receiving approval. A four-week queue and a fourteen-month queue have different implications to supply chain pressure.

Does an approval survive a model refresh? Robotics ships new SKUs annually. If approval is per model, the compliance cost recurs every cycle.

What this changes for you

If you build robots: pull country of origin on every BOM line before your next model cycle, starting with actuators. Ask each supplier which SKUs come off which line.

If you invest in the Physical AI stack: the exposed layers are the ones where the US line runs at defense or aerospace volume. Strain-wave reducers first, small-format cylindrical cells second, depth cameras third. Security certification is an open field, though it doesn’t move the Buy America component percentages.

Watch where Chinese component makers put new capacity. A shift toward US plants would be the first evidence that Conditional Approval is working as the onshoring bridge it claims to be.

If you run a research lab: models authorized before July 28 stay legal, so sub-$20k platform supply is now a depleting pool. Stock up on the units you’ll need for the next two years of research while the authorized catalog is still current.

Test out low-cost alternatives. Some examples are domestic and open sourced ones like Asimov by Menlo Research, Nori Robotics, Vibe Robotics. It’s to be determined if they pass the 65% test, and we have not found volume shipments from any of the three.

The rule reaches finished goods. The dependency it names sits upstream, in reducers and magnets and cells, where import stays free and the US lines that exist were built for aerospace and defense. Eighteen months from now we’ll know whether a gate at the border moved anything behind it.

Joining me is Jasmine Bai, Vice General Manager at FAIR Plus, where she runs international development. FAIR Plus is a Shenzhen exhibition co-organized by the Shenzhen Robotics Association and Messe Stuttgart, built for robot developers. Its second edition in April 2026 had more than 400 exhibitors and 60,000 visitors, roughly 3,000 of them international, across 100 countries.

Subscribe now

Where demand is strongest in Physical AI

Component suppliers took 70% of their exhibition floor. Full robot makers, 20-25%. It maps to the typical archetype of visitors for FAIR Plus: researcher-founders. A growing number of them also came for the brain side of the stack.

Visitor behavior differs by region

Europeans arrive having already concluded they need a hardware supply-chain partner in Shenzhen. Southeast Asian buyers come to source finished robots to resell into Malaysia and Thailand. American visitors, the largest international cohort, mostly came to observe.

We also covered a German robotics company relocating R&D to Shenzhen, the humanoid makers that drew the most crowded sessions from attendees, and the coordination layer above deployed robot fleets that Siemens is positioning to own.

Follow FAIR Plus here: https://fairplus.cn/en/about-fairplus/

You can also watch the show on Youtube:

Every episode, I sit down with the founders, researchers, and operators building physical AI worldwide, tracing the full stack from components and supply chains to what actually deploys. See all published episodes here.

Chapters
00:00 - Intro

01:41 - The strategic focus on robotic research and industry partnerships

03:03 - Surprising trends: American visitors’ prominence in China

04:45 - Regional motivations: Europe, Southeast Asia, and the US

06:52 - Insights into American visitor behavior and industry development

09:33 - The platform’s role in supporting robot developers and tech exchanges

12:07 - Types of visitors: universities, startups, and corporate innovators

13:36 - Academic researchers and their role in commercial robotics

14:34 - Trends in localization: German companies establishing R&D in Shenzhen

17:15 - Future directions: humanoid robots and industry adoption

19:14 - Popular sessions on collaborative research and supply chain innovation

22:23 - The increasingly unavoidable need to collaborate with China

23:17 - Notable exhibitors shaping future industry trends

24:23 - The open and collaborative industry environment fostered by companies like Siemens

FAIRPlus organiser Jasmine Bai on what American buyers actually come to Shenzhen for, and why component makers take roughly 70 percent of exhibitor count against 20 to 25 percent for whole-robot firms.

majority of them will agree that they have to work with China it’s not a question anymore the next question is about how to cooperate with China what about the Americans like what do they come to FAIRPlus for?

most of them just come and have a look components company take 70% 20 to 25 is whole robot 500 companies 60,000 visitors FAIRPlus is the largest physical AI exhibition in Shenzhen it covers the full stack hardware software and the supply chain behind it this year the most number of international attendees come from the United States Europeans and Southeast Asians came to buy hardware Americans came to talk Jasmine Bai runs the international side of FAIRPlus she sees who flies in and why we’ll find out what the world looks like from her seat Hi Jasmine, thank you so much for taking the time today yeah, thank you you run the international side of FAIRPlus the people who fly in the people who speak so can you walk me through how this job actually what you actually do and who are you trying to get into the building yeah, a lot of people will ask me because if I in charge of exhibition it’s only take three days per year what I’m doing the rest of the 300 days actually a lot of work beforehand to put everyone come at the same time and communicate for a specific topic for example for FAIRPlus we want to invite all the robotic related professors investors

robot developers to come to Shenzhen during three days and they can meet what the people they want to meet and talk with people they want to talk so actually my main job is go to study each market and try to find the key person invite them to come to Shenzhen and also let them to bring more people more companies or more friends to come together this year is the second year that you run FAIRPlus how many exhibitors, how many visitors and how many came outside of China so last year was our first edition of FAIRPlus we have 210 companies because we only have six months preparation so last year was majority from China side so from last April to this April we have whole year to do the international promotion so our exhibitor number grew to more than 400 and for our visitor is 60,000 visitors come to FAIRPlus and among is 3,000 are from international visitors and they are from 100 countries and a very interesting is number one is from America not number one is like top 10 American, Germany Singapore Korean, also Holland all countries are very focused on robotic currently

so when you saw the final visitor breakdown what was the most surprising thing for you yeah, actually it’s what I told you Americans visitors is really a lot because so far actually because FAIRPlus is co-organized by Shenzhen Robotics Association in the Messe Stuttgart, and Messe Stuttgart is a German company so we have a very good resource in European side and Shenzhen Robotic Association has a very good relationship in China about robot industry and also Messa Stuggart have very good office in Southeast Asia so actually you can know actually America is somehow we are not very good at so but in robots and AI industry actually America is very leading in this industry so actually from very beginning both of us said okay we apart from using our own very good resource we need to focus on American American side to do promotion and try to figure out how to influence the industry also we found it’s a little bit hard because it’s a very far from China to fly here and also the community is a little bit different and America had their own community to develop their own robot AI actually we work very hard but we don’t know the result so the result really surprised me that America is the No. 1 country from our international visitors part

so that also gives us a lot of confidence that in the future we can do more in this side and find out what’s their interest and how can I help them to do more you see where all these visitors come from each of them and when a European and Southeast Asian visitor come what are they here for and what about the Americans like what do they come to FAIRPlus for?

actually we know more about European Southeast Asia and European companies majority of them knows they need work with China for the supply chain actually when we do international promotion for FAIRPlus we highlight more about hardware cooperation because it’s non sensitive part and everyone should use it no matter you are do robot or AI you need some hardware and it’s no doubt that China is the best on manufacturing and production chain and also for robotics side actually Shenzhen is very good at that so everyone agree with the concept that they need to come to Shenzhen to find their partner for a hardware side and majority of Europeans very clear know they need supply chain partner here in Shenzhen that’s the key reason they come for Southeast Asia they want to buy more clear they want to some robot and bring back to like Malaysia, Thailand they are very interested to have some robot back to their country and resell it so it’s very clear the both part where the needs is so actually we do a lot of very matchmaking or targeted event to bring both side we select the Chinese company who can help the European or who can work with European side somehow work with Southeast Asian countries so we organize some small events to help them to matchmaking then they can go deeper to cooperate

but to be honest for American side so far we haven’t find some like very specific things even we have a lot of visitors I think most of them just come and have a look but it’s a good start after they come and have a look they understand the industry I think they will come up with more ideas what they need to do yeah, so it’s more of a educational tour for some of these visitors and they’re still exploring the use case and how to work with the Shenzhen partners you also mentioned that this year there are different pavilions and programs for different countries and there wasn’t an American program tell me more about that like what’s the story behind was the American conversation just happening in a different level I think markets are different like what I said Europeans, southeast Asian and Thailand they are more clear about their position in this industry how they can work with China of course some the majority of the idea from associations or our government partners or industry leaders from European or Southeast Asia they think in this way and they lead the topic of course not every company may follows but majority agree with this concept so they follow the rule to come and they benefit from it of course some of them also have different idea

I think American say they, I don’t have their own idea so far I also talk with the visitors and why they come and so first time and for the ones who first come to China they are super impressed by the how it’s working here because actually America also very good at robots so they have their own supply chain their own system to run it sometimes they don’t have to go abroad but when they come to abroad some idea come to their mind and maybe some from investment side also some hardware corporation or co-develop I’m also looking forward to more detail because our show is just happened one month ago so now we are still do internal review and communicate with some of the people who come to our show side and ask about their feedback and also what they want to do for this I remember you mentioned that FAIRPlus when you promote it internationally you usually focus on the hardware side but actually there’s a whole other side of the brain and the intelligence so tell me more about the full vision of FAIRPlus FAIRPlus actually is a platform designed for a robot developer so we welcome everyone who want to have their own robot or who already have their own robot because if they come they can see all the components all the OEM, ODM service company

and also software, hardware suppliers so they can talk with different part to realize his target to have the his own robot and what this robot use for because for different scenario the robots need to be designed in different way so actually this the whole concept for FAIRPlus is for robot developer so actually it’s a technical exchange platform not for sale or for market branding why is because so far like humanoid robot is not commercial yet even we see a lot of like Unitree, UBTech very famous they also need to upgrade their product it’s not ready that’s it can be sold to everyone or can apply also the other side is whether the robot is still finding more a scenario to be applied we think they can work in every area in the factory or in the home but we still need to define how to use it and also lower the cost so more people can use it so FAIRPlus also help robots to lower their cost because there are so many suppliers you can find out the best solution for your own robots then I really remember the association leader once say when the product the price of robots below enough there are more people will use it

then there the industry will grow will grow a lot yeah so actually so far we are focusing on technology exchange that’s why actually FAIRPlus only have shell skin every all the exhibitors are limited to 9 square meter or 18 so everyone is same and no matter you are Unitree or UBTech or Siemens or Bosch they all have very small booth so at beginning they don’t understand us but later we persuade them to do in this way actually we mean to do it because normally a visitor come to a show they will first go to the biggest booth we cannot change it because everyone is have the habit to go to the biggest booth but in our show we don’t have that every booth is the same so if you want to find your partner you need to go to all the booths so it’s somehow more fair to everyone in this market because actually it’s no top one or top two in this industry for the hands or robots they are only who already doing it but not the No. 1 this industry have a lot of future so we don’t want to limit it the visitors idea about which brand is good and also for the companies who come to our show they don’t need to prepare a lot of construction, decoration no, you just come to talk about technique

yeah so less marketing more like the technical innovations that’s happening and exchange learn from each other it’s very different so everyone come here with a different idea they don’t need to oh I should be looks better than my competitor no need you just come to talk with others about your technique I’m curious when you see people that come for the brain side of the equation to FAIRPlus who are these people?

are they universities or who are the buyers like universities or are they startups or are they big corporates that are looking to innovate in house yeah recently I just learn a new word that is a Researcher-founder and founder researcher it’s actually a very famous word in this industry especially in the data and brain side AI or brain side because actually it’s very the academy side world and industry very connected because you need for the data side you need a lot of research to find out how to control the robot and control each component and how did it come from lab to the real world so actually you can find a lot of startup founders they are from very high level universities or research or maybe they are professor their own so actually this part people are for who are the new founders for this topic so actually you see a lot of these researchers from universities to come for the brain side or the hardware side?

brain side but actually all the brain company also need hardware yeah, so they also need to find partner from hardware side so they need to work together it’s not only one people can do it they need partnership so actually FAIRPlus is best for them to find partners or find a solution for themselves yeah that’s also why we during FAIRPlus we have a lot of event one event is we do together with Science Robotics the Academy Journal from America you know the Science Magazine Science Robotic we do a session together with them want to link the academy side with the industry CTOs because from the research paper to a real company there is a lot of step need to go some researchers would like to go to industry or go to business and some still doing their research but they also need to know what the industry need then they do the research in that way so I remember when we chatted on the phone earlier you mentioned that you know a lot of times people think about Shenzhen is that oh we take a plane to Shenzhen find a partner and then get a supplier and you mentioned that there’s a company from Germany that actually set up their headquarters in Shenzhen so

tell me more about what you see in terms of the trend like are more people doing that now or is the former way still more popular?

yeah actually is a German company who do the robot hand and it’s called SCHUNK they also from Stuttgart their headquarter is from Stuggart we are very close with the company people and it was last year when we start our FAIRPlus they said oh we want we also doing this till now they’re main business is about supply the industry robots about the peak components actually this is their new business and last year their Global CTO will come to China to talk about us and also association to say that they want to build up a R&D center in Shenzhen about this department also they are sending their German colleague from technical like CTO or tech engineer to come to Shenzhen to lead this product because they think Shenzhen is a good place they can find their supplier also good people for the research side because Germany is a little bit behind the research or business in this now because Germany is very good at industrial robots but for humanoid robots they are a little bit late that’s why they come to China and they select to build up their R&D center here

I think more companies are considering it because you need to be close to the industry and also close to the supply chain so that you can know how to build that product I’m curious what do you see, you know given for example Germany that has decades or even a century worth of industrial robotics or industrial manufacturing expertise what is not transferable over to the robotic side, a humanoid side?

ideally in the future there’s everything can be transferred because for robots no matter human or robot the industrial robot is help us do our human’s things but one thing I can realize is from especially Germany side all the studies or the business is more focusing on humanoid robots application in the factories maybe upgrade the industry robot or use this robot to help the workers in factory they more focus on this side even I I read some research paper or they say okay the humanoid robot no need to have two legs because in the factory they don’t need to walk they can just move it’s very interesting I also talk with another speaker from American and I share the information oh I heard from Germany side they think the robot and then the American speaker tell me that oh if they don’t have they are not humanoid robot someone insists they should look at human right because the world is designed for a human being so look like a human is more comfortable to for there to walk maybe from cost effective side they have more ideas so apparently there’s no one answer about what the future should be so that’s how important

the communication platform should have I recently saw this video with a humanoid robots on skater wheels so they still have two legs but they’re on wheels so it’s like both two legs but also wheels so out of the FAIRPlus exhibition this year which part of the floor like which part of the exhibition has grown the most versus the first year? you mean the FAIRPlus?

yeah in terms of exhibitors yeah actually our percentage of companies are keep the same both edition this year and the last year our components company take 70% so like the actuators and 20 to 25 is the whole robot, whole machine actually the percentage is same almost same and we just grow the members of them actually this year still the international involvement is about speech and visit site so next year we are inviting more company to exhibit or take a booth here to show their things maybe from America or from Europe to have booth there to discuss with more people yeah and in terms of you mentioned that there’s a lot of speakers this year lots of different sessions what are the most popular sessions that you saw like the room is full and crowded with people?

we have survey on that actually the international exchange is the most welcome one like we have a European session, we have a very leading like a human robot company like superstar from Europe YSKL from Spain and one is Neura from Germany so both one CEO, one CTO to come here to introduce their product it’s very interesting for the Chinese company because what the first of what their robot are looking for where they can use why they have their how we can cooperate with them in the production chain because these two are still built up in Germany so their selling price is really high they also come to find out their solution because they know they need to lower the price and to benefit more industry to use yeah so the session from the Spain and is it German?

German German robots are they were the CTO and the CEO sharing about their supply chain like what were they talking about that attracted the attention? talk more about how their robots what they think the robots will be used and also how now also some painpoint of their company when they can cooperate with China because of course they come to China to find some partners they are not only fly here with 10 hours flight and come here just have a look it’s not so they’re looking for like cheaper components?

they will not say it but of course they are looking for that and so it’s European session we also have from Switzerland, Holland to somehow they have different very detail request but in general they want to cooperate with China and also we have another popular is Asia session is somehow same topic that is company from Japan, Korea and also another big catalogue very famous is about we have a conference called Power Robot is link the Capital Investment Institute with the robot company because now everyone need money also the robot company they are all a majority of them startups so they also want to talk with the investment and capital people to see so it’s like a investor and startup matchmaking yes platform so for someone overseas that came to FAIRPlus for a day what do you think that person will understand more after visiting FAIRPlus like about the China supply chain and how to work with suppliers in shenzhen I think majority of them will agree that they have to work with China it’s not a question anymore

some of them before they come they may be considered okay it’s a just a market but after they come okay it is definitely a yes to come to China the next question is about how to cooperate with China so different company have different ideas but they definitely will back to China again and also I got a lot like this space we do is as an extension of FAIRPlus I also got a lot of request how to come here to deeper to the industry so we are also working on it can you share two or three exhibitors that you would say hey if you come from overseas you should pay attention to these ones and why do you think those are up and coming?

is really a difficult question because actually why we put every booth the same side because we want they have a look at everyone but actually I want to share two companies who maybe if you want for German company I always take the example is SCHUNK they have their R&D center here not this one is the one I share and also like a big company Siemens they already work with us two years and their concept is to because there are so many humanoid robot and they need somehow a leader for the human and robot in the future in the factory so actually Siemens must position themselves in this position so they work with us and they talk with all the companies about their research and renting for Siemens they need to adjust themselves to know all the robot product how they can be used in the factory then they can be the leader or CEO of human and robot in the future is the leader also a humanoid or is it like I mean human humanoid because human need to control the humanoid robot and the humanoid need a leader to control the humanoid robot so Siemens want to be the part and they work with all the company so I want to share this is

they are very open to every company to talk and I think they don’t have business so far in this sector but they talk with everyone to understand different ideas so it’s very open to the industry then they can gain cooperation it’s not about which country you are come from it’s just cooperation yeah that’s awesome well thank you so much Jasmine, this is wonderful and very insightful so thank you so much for taking your time

Elon Musk has repeatedly called the hand the hardest part of Optimus, putting the forearm and hand at roughly 60% of the robot’s engineering difficulty. 72.8% of Unitree’s humanoids shipped in 2025 with no hand at all. This primer dives into why building a robot hand is so challenging, who is building it, how much it costs and the business around it.

Why Hands are Hard

The human hand mastered both the power grip and the precision grip in one organ. To replicate this dual mastery, it requires an intricate trade-off in the design.

To understand the robot hand, here are the key items on a hand spec sheet:

1. DoF - Degrees of Freedom: the number of independent ways an object or joint can move in 3D space. For example:

   1. 3 DoF is movement in three directions: X (forward/backward), Y (left/right), and Z (up/down).

   2. 6 DoF: an object can move freely in 3D space. This involves the three positional movements mentioned above, plus three rotational movements: pitch (tilting forward/backward), yaw (turning side to side), and roll (tilting side to side).

2. Number of joints: a joint is a place where the hand can bend; an actuator is a motor that drives the bending. The two counts differ in most hands, because one motor can drive a joint directly while other joints follow along through linkages or tendons. Vendors call the directly driven joints active and the follower joints passive. For example, Linkerbot’s L10 lists 20 joints, 10 active and 10 passive: 10 motors each drive one joint, and each driven joint drags a second joint with it through a linkage. The DoF-vs-actuation section below covers why buyers should count only the active ones.

3. Backdrivability: Fully backdrivable means when the joint is being manually moved, the motor easily spins backward. This allows machines to “feel” and react naturally to their environment, to safely interact with humans. The downside is that the motor must actively consume power to hold a static position. Non-backdrivable means the mechanism is mechanically “locked” in reverse. It’s ideal for holding heavy loads securely without using electricity, but because the system doesn’t comply when it encounters an obstacle, it can be dangerous for human-robot interaction.

4. Cycles: how many open-close repetitions the hand survives. Spec sheets almost always quote unloaded cycles, and 1-2M unloaded is the current top of the industry. In reality, hands are almost never unloaded. A teardown of 1X’s literature puts NEO’s tendons at ~2M cycles nominal but ~100k at 3x load - a 20x derating [teardown + company literature]. A metric to consider is the unit price divided by loaded cycles, that balances durability with the headline price.

5. Force: the spec sheets quote three different quantities

   1. Fingertip force, in newtons (N): what a single fingertip can push with — Sharpa 20N, NEO 45N, LinkerBot L10 15N per fingertip.

   2. Grasp force, also in newtons: the total squeeze of the closed hand, always the larger number. The range is around 45-100N. [confirm based on various datasheets]

   3. Wrist torque, in newton-meters (Nm): a rotational moment, not a force — NEO’s 17.75Nm cannot be compared to any grip force without specifying a lever arm. A vendor quotes whichever of the three flatters its design; the reader’s first job with any force number is to identify which one it is.

6. Weight: Both the hand itself and the forearm matter. Sharpa hand is 1.3kg, Wuji hand is <600g, LinkerBot L10 hand 800g. 1X’s Neo hand is unlisted (likely due to the high weight including the forearm tendons). A 500g hand rides a 3kg-payload arm with 2.5kg left for the task; a 4–5kg hand (eg. Shadow with forearm: 4.8kg) demands a 10kg-payload arm before the task gets a gram. Weight compounds up through the arm into total system cost.

A lot of these performance specs come from the drive architecture. Today there are 3 main types of architecture when it comes to building hands. They are all about choosing where, or whether to put actuators on the hands.

• Tendon-driven: The gears are on the forearms, instead of the hands. The cables (tendons) run through routing channels or pulleys, connecting the gears to the fingers and palms, using a pulling mechanism to make the fingers move.

• Direct-drive: Each joint is an actuator, independently moved.

• Rigid linkage: Actuators in some joints, and some in the palm. The other movements in the other joints are made through linkages

The key trade-offs are:

Tendon-driven hands allow for slim fingers and a lighter hand, with the weight shifted to the forearm. Shadow arm, for example, weighs 4.8kg with the forearm included. However the weak point is the anchor of the tendons. A 150N-rated tendon holds ~100N after limited cycles because the knot/sleeve termination degrades. It typically only works for thousands of cycles, not millions (sourced: Prensilia interview). As cables lose tension over time, movements also become less precise.

Direct-drive hands give full control of individual joints, one motor per joint. The sim-to-real gap is a clear advantage: there are no springs or tendon friction to model. The downside is that it usually comes at a cost premium as actuators are the most expensive line item in a hand. Another weakness of direct-drive design is thermal: holding a grip means motors continuously drawing high current, and small motors shed heat poorly.

Linkage hands are robust and cheaper than fully actuated hands as it requires fewer actuators. Weaknesses include (1) coupled motion, meaning the inability to manipulate individual joints, makes it impossible to do in-hand reorientation or piano-type gestures; (2) Grasp shape is determined passively by object contact, which sacrifices precision; (3) Compliant elements such as springs and linkages under load, are hard to simulate, lowering sim-to-real performance.

Failure modes (Mechanical)

1. Tendons lose tension or snapping: Repeated tension, bending, friction against routing pathways causes cables to stretch, fray or snap over time.

2. Fingers break: Fingers bump into fixtures, get struck or dropped on. They fail first because they take the impacts, hence quick-swap fingertips as a design requirement.

3. Gear wear, then failure: As the gear teeth slide and roll contact under load, and as lubrication breaks down or gets contaminated, it causes abrasive wear or cracks. This leads to a gap between mating teeth. The finger wiggles slightly even when the motor is locked, lowering precision and position control.

4. Motor overheating: Especially with the direct-drive design, in order to hold a grip in place, motors continuously draw high current, rapidly generating heat.

5. Gearhead structural failure at torque limits: Direct-drive or linkage hands use miniature gearheads, often with teeth at fractions of a millimeter wide. When a robotic hand accidentally hits a rigid surface (eg, smashes into a wall), the torque spikes past the gearhead’s maximum tensile strength which leads to the root of the gear tooth snapping off.

6. Wiring harness fatigue: the wires carrying power and signal to the finger motors and sensors flex with every joint movement, and copper conductors fatigue the way a paperclip does when bent repeatedly. Cracked conductors and loosening connectors produce intermittent electrical faults long before anything mechanical breaks. The flex life of the harness is estimated at ~1M cycles, which makes the wiring, on the company’s own numbers, the binding constraint on hand life [founder-claimed].

In addition, test conditions matter as much as the count. Each grasp type loads only some of the hand’s joints, so a million cycles of one power grasp proves out the three joints that grasp uses and leaves the rest of the hand untested [founder-claimed]. A cycle count on a hand spec sheet certifies the single motion that was tested, nothing more.

Reading the DoF number: Passive vs. active actuation, and abduction

Active vs. passive. If DoF is more than the number of motors, it is an underactuated hand, which means some of the DoF is controlled passively by a motor not at the actual joint but through linkage or tendons.

Abduction is also an important mechanism, which is the finger’s ability to spread apart laterally. It allows the hand to perform complex tasks like in-hand rotation, reducing task-completion time.

Flexion is the curl of a finger toward the palm, and it turns out very few independently controlled curl joints are needed. There are also studies in the research literature that basically say that having a hand that is able to have abduction movements and two to three degrees of freedom in the flexion of the fingers is equivalent to having a 20 degrees of freedom hand without abduction. When asked to spec a hand for a 10,000-unit humanoid, a researcher argued five motors can achieve ~90% of daily-living grasps.

The map: the hands market is a stack

The hands market today is growing rapidly. It spans the model layer, hand makers, components, integrators, and end applications.

Robotic Foundation Models: They build models for robots to operate in, increasingly adding in the dexterous layer on top of locomotion or pick and place. They shape the hand maker market in three important ways: (1) they buy or build hands for data-collection fleets (near-term merchant demand); (2) their architecture bets (vision-plus-proprioception vs. contact-rich training data) decide whether tactile is on the critical path; (3) their success decides whether the hand commoditizes into an interchangeable peripheral, with value accruing to the model layer.

Players: Physical Intelligence: hardware-agnostic VLAs from mixed embodied data. Generalist: GEN-1, trained on 500k hours of human interaction, adapts to a new body and task with ~1 hour of robot data [company-claimed]. Genesis AI: the vertical bet with its own hand, own glove, 1:1:1 kinematic mapping. RLWRLD: dexterity-first VLA with force/tactile channels, data from dispatched workers amplified by generative video [company-claimed].

Hand Makers: There has been an influx of hand makers, split into 4 lineages.

1. Pure-play merchants: Inspire, DexRobot, Proception, Shadow, Wonik, Psyonic, Prensilia);

2. Humanoid OEMs building captive: Tesla, 1X, Figure, AGIBOT, Unitree);

3. Actuator/component companies moving up-stack: the Maxon/Faulhaber-adjacent path, Wuji, LinkerBot, and Zhaowei;

4. Research platforms: Allegro, Shadow DEX-EE with DeepMind, Orca with ETH Zurich (tendon driven).

Components: key components inside a robot hand are

• Actuators: For the direct-drive or linkage based design, it requires miniature actuators, which requires high precision gears. It is a very specialized production line, which is why many actuators companies are starting to make hands.

• Encoders: Actuators require ultra-precision encoders, a lot of which are sourced from the West.

• Tendons: the cable itself is Dyneema (or the equivalent Spectra), an ultra-high-molecular-weight polyethylene fiber stronger than steel by weight, sold by the spool as a fishing-line and rope commodity by Avient and Honeywell. The fiber is not a bottleneck: it is cheap, abundant, and not hand-specific. The engineering value sits one layer up, in termination and routing.

• Tactile sensors: There are different designs behind tactile sensors which we’ll cover in the next section, and there is no industry standard in tactile sensing.

Tactile: the twin problem

Hands allow humans to interact with the world, and through manipulating objects, we gain understanding of the world around us. The hand is an active sensory organ: touch isn’t a passive input channel but an experiment the body runs continuously. In order to capture contact information, tactile sensors need to be placed on the robot hands. Tactile, as it turns out, is a separate unsolved stack with its own frontier.

Tactile has no standard. The supply base is startups and academic spinouts with no production-scale manufacturing yet, globally. The tactile is a younger industry than the hand itself.

Overview of Tactile Sensors Providers

Below is a list of players: vision-based (GelSight), magnetic/Hall, capacitive skin arrays (XELA). Like hands, durability is a metric that’s not listed on the spec sheet. The deployment bar is around 5 million touches, where lower-cost sensors age out at 1 million.

Another approach is putting a tactile glove onto the hand, where all sensors are on the tactile glove and the hand makers focus on the mechanical aspects of it. The glove acts as the bridge between tactile hardware and manipulation data.

Is tactile prerequisite or luxury?

The deeper question is whether manipulation data must include contact, or whether vision plus proprioception can substitute. The industry is making different bets.

Wuji ships encoder-only (”in the end state the glove is the only solution”). Many hand makers choose to outsource sensing entirely. A Shenzhen micro-drive manufacturer we interviewed specs and sources its tactile, rather than building it. 1X puts it in the glove material, not the skeleton. In contrast, Tesla put palm sensors on its Optimus V2 hands.

The physics behind shipping without tactile sensors is transparency. In a finger with low gear, the motor and fingertip stay mechanically linked in both directions. So pushing on the fingertip shows up as a change in the motor’s current. This lets the controller estimate fingertip force from motor current alone, with no force sensor in the finger.

The strongest evidence for “not on the critical path” is what the model layer omits: Levine’s recent talk on dexterity contains no tactile sensing at all [academic, absence datapoint]. HIL-SERL reaches 100% success on chip insertion from images alone in ~15 minutes of training, 2–3x faster than imitation baselines [academic]. And RLDG finds RL-generated data outperforms human demonstrations at equal success rates [academic]. Vision plus proprioception keeps clearing bars it was supposed to miss.

The Retargeting Gap

Glove-collected human motion has to be retargeted into robot command space, and there is no consensus method: force feedback is not touch, and “if the data is junk, the robot’s going to be junk”.

Researchers are also testing the “train-rich / deploy-lean” hypothesis: dense tactile during the teaching phase, simpler or no sensors at deployment. If this holds, it may limit tactile demand to only training fleet and not in mass deployment.

The cost curve: the BOM and which component costs are collapsing

Assembled-hand prices are falling fast: $100k-class research hands, to $20k, to Inspire’s $3–8k at roughly 10,000 units delivered, toward a claimed sub-$1k floor.

The key cost drivers of a hand are: precision gearboxes (the actual cost driver), matrix tactile sensors (disproportionately expensive per unit vs load cells), and the operating-model gap between design firms and volume manufacturers.

BOM of a hand

The BOM shape depends on which supply chain builds the hand, and the difference is the point:

• In a European build on catalog motors, mechanical transmission and structural frames run ≈70% of BOM, and the motors alone are roughly one-third of total cost [company-claimed, Prensilia].

• In a Chinese build, the motor is the cheap part: brushed motors at RMB 50–100 (US$7-14), brushless at RMB 500–1,000+ (US$70-140), an order of magnitude apart. The precision gearbox is the expensive stage of the actuator [call, anonymized].

• PCBs and electronics are already cheap and fall further with scale.

• Tactile sensors stay disproportionately expensive per unit relative to load cells.

The motor line item collapses as sourcing moves to Chinese vendors. The precision gearbox resists collapse in both geographies, which is what makes it the cost driver.

The challenge is not just the BOM but also manufacturing scale.

For research grade makers, the hands are in-house engineered, outsourced fabrication and in-house assembled. The volume tops out at hundreds of hands per year. Going to thousands per year, requires redesigning for volume processes, partnering with assembly firms.

For players that come from motor and micro-actuators backgrounds, their already-available mass production capability enables them to scale quickly. One listed maker is expanding hand capacity ~70% in H2 2026 at its existing plant.

And that is why Inspire is delivering 10,000 units at $3–8k and LinkerBot claims 80% hand market share with its sub-$5k hands.

Beyond purchase price, the total cost of ownership (TCO) is a key metric to compare to evaluate ROI. That includes maintenance, reliability and uptime. Tendon hands need periodic re-tensioning, teardown analysis expects 1–2 year replacement cycles on wear components, and sensor gloves are consumables that tear [teardown, 46:39]. Direct drive designs have overheating issues; sustained grips draw continuous current, limiting uptime.

What exactly is a hand business: Where value sits

Is a hand company a data company with hardware, or a component supplier with component margins?

The revealed volumes are small. Unitree discloses Rmb 17.2mn (US$2.5mn) of Inspire hand purchases over the first nine months of 2025, for 1,210 hands, 96% of everything Unitree bought in hands from outside suppliers [filing]. That is a pilot-scale line item for a company shipping humanoids in the thousands. Unitree is also building its own Dex5 in-house [filing]. Sharpa claims mass production and has published no numbers [company-claimed].

For many merchant hand makers, the first market is university and corporate research labs. Unitree’s early revenue came from selling robots to academic labs, which proved that academia buys robot hardware in real volume. Wuji, LinkerBot, and a few other Chinese hand makers picked the research first approach. The incumbents in that field are Inspire and Shadow.

The third paying channel is industrial automation, and for the Western merchants it is already billing. Prensilia CE-marked the Mia hand as an industrial gripper in early 2025 and puts its buyer mix at 50% research, 35% prosthetics, and 15% industrial, the segment it is now pushing hardest [company-claimed]. The lead use case is human-machine-interface testing. Automotive plants station a person inside each car at the end of the line to test the interior, and a lightweight hand on a standard robot arm can run the same touchscreen, button, and lever checks; ultrasound machines and other medical-device interfaces follow the identical logic [company-claimed]. PSYONIC reports the same pattern from the US side: over 50 robotics customers, including NASA, Meta, Google, and Amazon, mount the Ability Hand on humanoid robots and on industrial robot arms, for tasks that run from car assembly to laundry folding [founder-claimed].

The framing that fits this revenue is the hand as the flexible complement to the industrial gripper, riding arms that are already deployed. Robotiq and Schunk are the incumbents whose customers these dexterous hands sell to: the gripper wins single-part work on speed, force, and repeatability, and the hand wins the cell that handles many part types or an interface designed for human fingers [company-claimed]. Humanoid programs supply the category’s headlines. The invoices, for now, come from installed arms.

Platform or component

Not every hand can be a standalone product. Forearm-integrated designs (Tesla, 1X, Wonik’s Allex) cannot be unbundled and sold separately: the actuation lives in the forearm, so there’s no product boundary at the wrist.

For the hands that can be sold, the question is what keeps a customer.

1. Data: Training data collected on one hand is recorded in that hand’s kinematics. That does not survive a switch to another model.

2. Code: Customer’s software stack is written against the vendor’s SDK: the control APIs, the simulation models, the calibration routines, the data pipeline. Swapping the hardware takes an afternoon; rewriting that integration, re-tuning controllers, rebuilding the sim environment, and re-validating everything downstream takes months.

Together, choosing a hand is closer to choosing a platform than buying a component.

That is the logic of the cheap-hand strategy: make the cheapest hand that clears the performance bar, win the research market on price, and let dataset and integration code accumulate on your kinematics. The equilibrium price then becomes the marginal cost of its motors, plus a standard SDK intended to become the default research platform.

Where value sits in the hand stack

Currently humanoid makers are split in the approach: a) building in house for more control, or b) buy for speed, and often, buy expertise in reverse engineering and time for internal development.

When bigger humanoid players make their hands in-house, they still buy parts. The squeezed party is the finished-hand merchant, who loses the order when the customer goes captive. What remains for merchants is the software-led programs, and there the defensible asset is the switching cost described above.

Does the model layer or the hardware capture value for robot hands?

Leading foundation-model players such as Physical Intelligence and Generalist are hardware-agnostic, built to treat any robot embodiment, hand included, as an interchangeable peripheral. If that works, hand makers compete on price like component vendors. The scarce asset is then the model and its data.

The flip side is, if manipulation performance stays tied to hand-specific kinematics and hand-specific data, then the hardware layer keeps its value, as the product is the hand, glove and model all tied together.

Every hardware-agnostic success moves value up the stack, every vertical-integration success holds it down.

If hands do commoditize, the unglamorous industrial layer then becomes the gating factor: system integration, reliability and the test protocol. A test protocol is a reproducible procedure for proving performance claims: which grasp, at what load, at what temperature, for how many cycles, with what counted as failure. The vendor whose test protocol buyers trust ends up writing the rules of the comparison. Semiconductors already work this way: datasheet conventions began as individual vendors’ formats and hardened into the shared language every part is judged in. The hand market is one credible protocol away from the same structure.

And whether the dataset is the moat is to be determined. There are at least three ways collected data could fail to improve policies:

• Models may generate better data themselves. RLDG finds RL-generated data outperforms human demonstrations at equal success rates [academic]. Every result like it shrinks the value of a stockpiled demonstration corpus.

• The recording may lose what matters in translation. Glove and teleop recordings must be retargeted into the robot’s command space, and Section 3 showed that translation degrades exactly the contact behavior training needs. Teleop quality compounds the problem: “if the data is junk, the robot’s going to be junk” [teardown commentary].

• Deployment may out-produce collection. Working fleets generate manipulation data as a free byproduct of paid work: PSYONIC’s prosthetic hands, in daily use on 300 patients, are a data source no glove-collection operation matches on realism or cost [founder-claimed]. If deployed fleets scale, purpose-collected data becomes a wasting asset.

The selection framework

Read more

This year, dancing robots are no longer the spotlight. What stole the show was how many hours a robot runs on a real line, and at what success rate.

Xiaomi puts numbers on the table

After four months on its own EV assembly line, Xiaomi's Robotics-0 model took a self-tapping screw station from 90.2% to 98% success, roughly a point off the human benchmark, and picked up two new tasks at around 90%. It is a yield number in a real factory, not a demo figure in a lab. That's the kind of metric that matters to a plant manager.

Xiaomi is a unique player in the Physical AI space because it owns the factory its robots deploy into. Earlier last week, on July 15, it also open-sourced U0, a 38B multimodal model, and topped the WorldArena leaderboard.

Xiaomi showed the CyberDog quadruped in 2021 and the CyberOne humanoid in 2022, and released the Robotics-0 VLA in February 2026. This past week was the first time it has put hard factory-yield numbers on the table.

Factory deployment is the main character

The common theme across the booths was deployment evidence on real factory lines. Companies shared runtime and first-pass yield, the metrics buyers actually underwrite. Below is what we heard (all figures are company-claimed):

• AgiBot G2 Max into a JD Logistics warehouse: 24-hour palletizing at 18kg per arm, framed as the first humanoid on a live warehouse production line.

• Pia Automation (均普智能): multiple robots assembling BMS (Battery Management System) units at 99.9% success, on a 70-second cycle.

• Spirit AI (千寻) Moz1, its general-purpose humanoid: on a CATL battery-pack line, reporting above 99% on high-voltage plug insertion. It also debuted Moz2, and framed its strategy as “industrial first, then commercial, then home.”

• Galbot: units have run 7×24 at CATL for three months, and in pharmacies for over a year.

The brain race moved toward production

The model layer is consolidating around VLA and world models. The focus is on being deployable rather than demonstrable, a shift from capability claims to cost-and-reliability claims.

• Unitree UnifoLM-OminiA-0.3: released July 20, a single model running home-care tasks on the G1 with no switching between separate perception, decision, and execution models, which Unitree says lowers onboard compute. It was shown via a demo video (tidying, pill-box sorting, loading a dishwasher), rather than live deployment. Unitree is the shipment leader in 2025, more than 5,500 humanoids and roughly 11,000 G1 units off the line by May 2026.

• AgiBot: a three-part stack, GO-2 (action), GE-2 (which it says won the 2026 World Arena world-model track), and Genie Evolver (an RL layer). AgiBot is trying to own the whole intelligence layer on top of its own hardware, and the world-model piece is what governs whether a robot generalizes beyond trained tasks.

• Ant Lingbo: one brain driving mixed-vendor robots through a live pharmacy workflow. LingBot-VLA reportedly drove ~20 robot morphologies from 17 vendors, and completed a pharmacy fulfillment task in ~90 seconds. The cross-embodiment angle is the interesting part: a bid to be the software layer across many bodies rather than one robot's brain.

• AI2Robotics NeuroVLA: a brain-structured model following a cortex/cerebellum/spinal analogy. It sounds compelling, but the details were thin.

• WeRide WITT: a physical-AI model it says cuts token cost by up to 98% (company-claimed).

Capital pooled into the data and supplier layers

• LightWheel (光轮智能): pure-play embodied data and simulation, valued around ¥15bn (US$2.1bn) with Ant leading the round. It called itself the category's first unicorn and cited Q1 2026 orders of ¥550mn (~US$76mn).

• LimX Dynamics: roughly $200M pre-IPO at a ¥15B valuation. Lens Technology, an Apple supply-chain name, is one of the investors.

• Lens × Swancor (上纬新材): a first-phase 10,000-unit line of finished robots. Lens supplies actuators and joint modules to AgiBot and does whole-machine assembly for AgiBot's Lingxi (灵犀) line. Lens also cites 500k units/yr of dedicated capacity.

• 麦麦科技: orders for 20,000 robots across 16 agricultural scenarios in Malaysia.

Components matured in public

The less visible half of the deployment story is in the subsystems, where domestic suppliers are showing up across perception, actuation, and control.

• RoboSense E2: solid-state perception on a self-developed SPAD chip, 3x the prior precision.

• Joyson (均胜电子): a 20-DOF hand (16 active), 380 Wh/kg solid-liquid battery, denser actuators.

• Oymotion (傲意): over 10,000 dexterous-hand orders in the first half. UMA Robotics’ recent debuted humanoid is one of their customers.

Read more

On Friday evening July 17 PST, a rumor circulated on X: Anthropic is pursuing Physical Intelligence.

Neither company has confirmed it as of Monday July 20 11:30AM PST.

Physical Intelligence was founded in 2024 by Karol Hausman (CEO, ex-Google Brain), Sergey Levine (Berkeley), and Chelsea Finn (Stanford), among others. It raised $1.1bn across three rounds, valued at a $5.6bn in November 2025. It was reported in talks for $1bn more at $11bn in March 2026. The company never confirmed the latest round publicly. It has no disclosed revenue. Pi is founded on the vision that intelligence, not hardware, is the binding constraint for robotics, and that one model can generalize across any robot body. The company blog has not been updated since the π0.7 release in April, after seven posts between November 2025 and April 2026.

If Physical Intelligence is selling, the likely need is capital, not data or end customers, since Anthropic operates no fleet and cannot supply robot interaction data. If Anthropic is buying, it is building an embodied AI position, following OpenAI’s in-house team and Google’s Gemini Robotics group. Either way, the deal reads as capital and strategy, not data.

The model-layer approach faces three structural challenges:

The embodiment gap. π0.7 showed real progress on zero-shot compositional generalization and self-correction. But the tolerance for error in the physical world is unforgiving: 90% in a lab is not the 99.999% that commercial deployment underwrites. “Grasp the object” resolves to entirely different torque, joint-velocity, and compliance profiles depending on whether the actuator is a harmonic drive or a planetary gearbox, and whether the arm has six degrees of freedom or seven. Much of that hardware variance originates in the Japanese and Chinese component base most Western model labs never touch directly. A hardware-agnostic model still requires bespoke per-platform fine-tuning.

The deployment-data bottleneck. Foundation models improve on fresh tokens, and in robotics the valuable tokens are real-world: edge-case failures, surface variation, tactile feedback. A hardware-agnostic company does not own the environment that generates them. Its partners do. Weave, running π0.6 folding laundry in San Francisco laundromats, cut missed grasps 42% and interventions 50% once Weave’s deployment data was included in pre-training. Ultra, packing e-commerce orders across a US warehouse fleet at 96.4% autonomy on a full shift, saw throughput rise the same way, by feeding Ultra’s data back in. The structural point is legible in Pi’s own numbers: the model improves fastest when someone else’s robots feed it.

The question is how much of the value the model layer can keep from data its partners generate and own.

Skild, on the other hand, embeds directly in industrial lines, running on NVIDIA and Foxconn floors. Tesla and Figure sit at the far end, owning the body outright.

Liability and safety. Safety is the gate to scaled deployment, at home or on a line. When a robot injures a person or damages property, attribution is genuinely hard: a spatial-reasoning error in the model, a latency spike in compute, or a mechanical failure in the hardware, and which triggered which. Until that chain is underwritable, corporate buyers hold back from mass deployment.

Read more