Physical Intelligence unveils OM-1: robot foundation model trained purely on human data, zero teleop
zipengfu · x · 2026-09-15
Physical Intelligence has introduced OM-1, its first robot foundation model, zero-shot generalizing to any robot embodiment: tabletop arms, industrial arms, and humanoids.
Key claims: it learns directly from human manipulation data with no teleop or robot data, achieves close-to-human-level dexterity and efficiency, and supports multi-robot collaboration.
Robotics researcher @mangahomanga (retweeted by zipengfu) called the results "very impressive," noting the lead author's PhD papers on manipulation from human data were among his favorites — lending credibility to the human-data scaling approach in robotics.
More from Embodied
- Hackathon Project Skillio Turns Vision Pro Into an AR Screw-Driving Coach Agent — seanmcdonaldxyz · 2026-09-15
- Autonomous robot unloads trucks without any fixed infrastructure — TinfoilTricorn · 2026-09-15
- Tesla FSD handles full descent on Slovenia's highest road at 2,055 m — FlorianGallwitz · 2026-09-15
- Physical Intelligence unveils OM-1, a robot foundation model trained on human data — kenziyuliu · 2026-09-15
- Unitree's new dual-axis robot neck mocked: tilt-then-pan kinematics force absurdly long neck — carlosdponx · 2026-09-15
- Reward AI debuts Omnibody Hand & OMI-1, pushing hardware-algorithm co-design for cross-embodiment dexterity — zipengfu · 2026-09-15