Skip to content
OrganisationsReward AI

Reward AI

Software vendorResearch
Commercial
Followers
0
Sign in to followVisit website
About

US robot-learning company building a general-purpose manipulation policy that is trained on human demonstrations rather than teleoperation or on-robot data. Its first model, OM-1 (Omnibody Model 1), is captured through a wearable 7-DoF sensorised hand and is designed to run across robot arms, legged humanoids and wheeled mobile manipulators from one shared data interface. The team came out of stealth in September 2026 and grew out of the Stanford robot-learning work behind DexCap, HumanPlus and ALOHA. No weights, code, dataset or API have been released.

At a glance
  • Latest product
    OM-1 (Omnibody Model 1)
    3w ago
Products
View all (1)
OM-1 (Omnibody Model 1)
Grasp / Manipulation Model

General-purpose manipulation policy trained only on human demonstrations captured through a wearable 7-DoF hand — no teleoperation and no on-robot data — and designed to run across arms, humanoids and mobile manipulators.

Activity
Recent
Links