OM-1 captures subconscious human physical intelligence directly from people, building on Stanford's DexCap
zipengfu · x · 2026-09-15
Reward AI explains OM-1's data source: the model captures subconscious human physical intelligence directly from people—nudging, sliding, twisting, power grasping, and whole-body coordination. Data collection builds on DexCap, the team's prior Stanford work: a portable hand mocap system using SLAM and electromagnetic tracking, paired with the DexIL imitation algorithm that turns human hand mocap data into dexterous robot skills, with an optional human-in-the-loop correction mechanism.
More from Embodied
- RewardAI unveils OM-1 robot foundation model trained from human manipulation data, no teleop — zipengfu · 2026-09-15
- AgentSTAR tracks and reconstructs articulated objects from monocular video through occlusion — notmahi · 2026-09-15
- Valve launches Steam Frame wireless VR headset in 256GB and 1TB models — Scobleizer · 2026-09-15
- OM-1 robot and Omnibody Hand debut, packing tacit dexterity into one model — zipengfu · 2026-09-15
- Fastest robot manipulation model yet, learned purely from human data — zipengfu · 2026-09-15
- RewardAI unveils OM-1 robot foundation model trained on human data only, zero-shot across robots — zipengfu · 2026-09-15