Reward AI

Reward AI builds general-purpose robot intelligence for any robot body. Its flagship OM-1 (Omnibody Model 1) is a single robot policy learned firsthand from human manipulation data — captured via the Omnibody Hand wearable — that deploys across industrial arms and humanoids at human speed.

Most robot-learning pipelines still lean on slow, expensive teleoperation and on-robot data collection; Reward AI's bet is that natural human dexterity, captured directly, is the faster route to scalable manipulation. It exited stealth on 14 September 2026 with a stack built around one model, one data interface and any body.

The company launched in 2026, founded by Chen Wang, to attack a stubborn bottleneck in robotics: collecting dexterous manipulation data without a robot in the loop. The premise is that learning from humans firsthand, rather than teleoperating machines, is the more direct path to general-purpose control.

Details on the founding team's pedigree aren't public yet, but the technical thesis is unusually specific — a wearable data interface, a shared cross-body policy, and human-speed deployment. It's the kind of end-to-end stack that suggests deep conviction about how robot intelligence should be built.