RoboBoom
The Humanoid Hub· September 10, 2025· 54:47

“You Don’t Scale a Crappy Robot” - Scott Walter

Scott Walter, a robotics expert, joins The Humanoid Hub host to analyze Tesla Optimus hardware, speculate about Gen 3, and discuss production, training data, and scaling.

In this talk

  1. 2:44

    Why the wrist and hand need integrated design

    Walter explains how Optimus Gen 2’s parallel wrist mechanism enables strong pitching motion but restricts side-to-side movement. He says the design can also leave space for routing tendons through the forearm, which matters when designing the wrist and hand as a connected system.

  2. 7:00

    Gen 3 hand and actuator trade-offs

    Walter describes how Tesla appears to have repackaged the wrist and hand, including smaller linear actuators and a revised wrist linkage. He says the 24-DoF hand may still present challenges in grip strength, finger spreading, tendon control, and teaching the robot to grasp. He suggests a hybrid approach could use tendons for finger flexion and linkages for spreading.

  3. 14:33

    Supply chains depend on design choices

    Walter says Gen 3 may change much of Optimus’s actuation strategy, but Tesla has not clarified whether it will favor rotary or linear actuators. The two approaches require different equipment and suppliers, and specialized production machinery can take six to twelve months or longer to obtain. He says late design changes could leave suppliers with capacity they no longer need.

  4. 34:03

    Training robots with human video

    Discussing reports that Tesla is emphasizing human video over motion-capture suits and teleoperation, Walter notes that Tesla had demonstrated first-person data collection earlier. He says workers should learn a consistent procedure before recording tasks, so the training data reflects repeatable methods rather than each operator’s personal habits. He also frames teleoperation as a possible way to test hardware capabilities while robot training methods develop.

  5. 30:55

    Scale after the robot is useful

    Walter says Tesla may have reduced Gen 2 plans to avoid scaling a design that did not meet its goals. He estimates production could reach around 1,000 to 2,000 robots in 2026 if Gen 3 prototypes and supply-chain ramp-up go well, with 10,000 possible in 2027. He says a million by 2030 is possible, while expecting near-term production numbers to disappoint and emphasizing that factory work still needs to prove useful.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks