The fight over humanoid hand design has spilled out of the labs and into the open. Brett Adcock, CEO of Figure, called the company’s 2022 tendon-based gripper one of his worst engineering decisions in four years. The rebuttal from rivals was immediate and sharp.
Key takeaways
- Brett Adcock (Figure) called the F.01 robot’s tendon hand “one of the worst engineering decisions” and a “complete local maximum”
- Andrea Esposito (Foundation) dismissed the claim as a “skills issue”, not an architectural flaw
- Figure’s early hand had only 11 degrees of freedom (DOF) — too few for tendons, according to the critic
- Tesla announced 25 actuators in the Optimus Gen 3 hand (up from 17 in the previous prototype)
- 1X uses tendon drive in its NEO robot: 22 DOF per hand, focused on safety and compliance
Two schools of hand design
The industry has split into two camps. Direct drive?direct drive: A design that places miniature motors directly in the finger joints — no tendons or intermediate transmission. puts miniature motors straight into the finger joints. Tendon friction disappears and control becomes more predictable. The cost is a hard ceiling — finger volume limits motor power and heat dissipation.
This side has real deployments behind it. Sharpa mass-produces vision-based tactile hands with 22 degrees of freedom. The direct-drive platform from Wuji Tech was praised for exceptional robustness and high predictability.
Tendon drive works differently. Heavy actuators sit in the forearm and force travels to the fingers through tendons. The hand gains human proportions and more power. Foundation showed an open-loop tendon hand catching baseballs — with no touch, relying only on TMR (Tunnel Magnetic Resistance) sensors?TMR sensor (Tunnel Magneto-Resistance): A tunnel magneto-resistance sensor — reads position and force from changes in a magnetic field, with no mechanical contact. and Jacobian-matrix estimators.
| Dimension | Direct drive | Tendon drive |
|---|---|---|
| Actuator placement | Miniature motors straight in the finger joints | Heavy actuators in the forearm, force via tendons |
| Dexterity / DOF | Sharpa 22 DOF, AGIBOT Ultra-M 20 DOF | 1X NEO 22 DOF; Figure’s early hand only 11 DOF |
| Power / force | Power ceiling — finger volume limits the motor and heat dissipation | More power thanks to large forearm actuators |
| Controllability | High predictability, no tendon friction | Tendon friction, especially Bowden tubes, hurts control |
| Tactile sensing | Vision-based tactile hands (Sharpa, Wuji Tech) | Foundation demo: open-loop grasp with no touch, on TMR sensors |
| Safety / compliance | Not highlighted in the dispute | Natural compliance and safety (1X NEO) |
| Deployments | Sharpa, Wuji Tech, AGIBOT OmniHand 3 Ultra-M | Foundation, 1X NEO, AGIBOT OmniHand 3 Ultra-T, early Figure F.01 |
A public exchange of blows
Adcock called tendons a “complete local maximum” and announced Figure’s move to a seventh-generation hand with finger abduction and adduction. The counterpunch came from Foundation. Andrea Esposito, mechanical design lead, framed Figure’s failure as a “skills issue”.
Esposito pointed to specifics. Figure’s early hand had only 11 DOF — too few, in his view, for a tendon system. He also criticized the use of Bowden tubes?Bowden tube: A flexible cable-in-housing that transmits force through an inner wire; it adds heavy friction that degrades precise control., which create heavy friction and hurt controllability. As a counterexample he cited 1X, which uses tendons for compliance and safety.
Multi-strategy players
Not everyone picks a side. AGIBOT (AGILINK) makes both variants — the OmniHand 3 Ultra-M with a 20-DOF direct drive and the tendon-based Ultra-T. Tesla, in turn, is raising the actuator count in the Optimus Gen 3 hand to 25.
A separate dispute over Foundation’s credibility runs in the background. Blogger Mike Kalil questioned the lack of independent confirmation of deployments in Ukraine and of a 24 million USD military contract, as well as a claimed 100 million USD annual recurring revenue.
Why it matters
The hand is the hardest part of a humanoid and the bottleneck for the whole industry. The architecture choice determines what a robot can actually do — grip gently, carry loads, or stay safe next to a human. The public clash between CEOs shows there is no winning standard yet. Each approach has a different physical ceiling. For manufacturers the risk is real — betting on the wrong architecture sets a project back years, as Adcock admitted outright. The market is only now converging on a dominant pattern.
What’s next?
- Figure is rolling out a seventh-generation hand with finger abduction and adduction — a test of Adcock’s thesis on the superiority of direct drive
- Tesla is set to show the Optimus Gen 3 hand with 25 actuators, which will test the scalability of its approach
- Foundation must provide independent confirmation of the contracts and revenue publicly questioned
Sources
- Humanoids Daily — The Great Dexterity Divide: Why Robotics Leaders Are Warring Over Tendon vs. Direct-Drive Hands
- 1X Technologies — NEO
