Robotics researcher & builder working on physical AI, robot identity, and cultural/economic infra for autonomous robots. Prev NASA, Goldman Sachs, Mysten Labs
Impressive. It’s 1X speed. Not a VLA model btw.
Training described as “learning robot actions directly from human motion, no teleop and no on-robot data”.
So closer to a scaled, embodiment-agnostic imitation policy (visuo-tactile-proprioceptive) than to OpenVLA / π0 VLAs.
Introducing OM-1, our first robot foundation model, zero-shot generalizing to any robot: table-top arms, industrial arms and humanoids.
- learned directly from human manipulation data
- no teleop/robot data
- close to human-level dexterity and efficiency
- multi-robot collab