inFollow
inFollow
Reward AI

Reward AI

"Frontier intelligence for any robot"

Software/Algorithm

Bay Area robot-learning startup founded 2025; its OM-1 policy learns manipulation only from human demos captured with a 7-DOF wearable hand.

Visit Website

Company Overview

Reward AI is a robot-learning company in the San Francisco Bay Area that develops a general-purpose robot policy trained on recordings of people rather than on robot data. It was co-founded by Zipeng Fu, chief executive, and Chen Wang, chief technology officer. Fu completed a computer science PhD at Stanford's AI Lab under Chelsea Finn and previously worked as a researcher at Google DeepMind; Wang completed a Stanford computer science PhD advised by Fei-Fei Li and Karen Liu in 2025, with research stints at Google DeepMind, NVIDIA Research and MIT CSAIL. The company's LinkedIn profile gives 2025 as its founding year, and the team describes its background as robot learning for dexterous manipulation, mobile manipulation and legged locomotion.

The company came out of stealth on 14 September 2026 with OM-1 (Omnibody Model 1), the decision-making core of what it calls the Omnibody stack. OM-1 is trained only on human demonstrations captured with the Omnibody Hand, a wearable seven-degree-of-freedom capture device that the company traces back to the team's earlier DexCap research system. No teleoperation and no on-robot data are used, and pre-training and post-training are merged into a single stage. The model takes camera images, tactile signals, inter-finger proximity and hand-pose trajectories, each at its sensor's native rate, with force recorded along the trajectory, and outputs motion direction, speed, force and the timing of events such as grasping. A separate low-level controller, trained with reinforcement learning in simulation, runs at high frequency and converts policy actions into robot motion.

Reward AI states that a single OM-1 policy runs across different robot bodies, from industrial arms to humanoids, and on mobile robots for navigation. Demonstrations released with the launch show conveyor-belt sorting, opening a fully closed refrigerator door, lifting delivery boxes of varying weight, and dynamic motions such as tossing and swinging, presented as fully autonomous at 1x speed. The company claims a new task, including long-horizon and dynamic ones, can be learned from less than 30 minutes of human data. For the capture hardware it reports that an electromagnetic tracking approach cut mean overshoot error by 60 percent at high speed, from 24.9 mm to 9.5 mm at 67 cm/s.

As of October 2026 no model weights, code, dataset, API, paper or benchmark results have been released, no customers or pilots are named, and no funding round or investors have been disclosed. The company publishes a general contact address and a careers address, and its website is limited to the OM-1 announcement, an about page and a blog.

Capabilities & Activities

Primary type & automation activities this supplier delivers:

Applications & Industries

Product Categories

Products & Solutions

OM-1 (Omnibody Model 1)

Robot policy trained only on human demonstrations, stated to run across arms, humanoids and mobile robots.

Omnibody Hand
  • Launch Year: 2026

Wearable 7-DOF hand-capture device used to record human demonstrations for OM-1 training.

Partnership & Notable clients

Associating Events

Listed in RobotToday Supplier Discovery. Verified Profile

More related news

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20

Contact Reward AI

WEBSITE

https://www.rewardai.com

EMAIL

[email protected]

HEADQUARTERS

San Francisco , California  94103

United States

Company Facts

Founded

2025

Primary Role

Software/Algorithm

Company Size

-

Primary Region

North America

Annual Sales

-

Funding Stage

-

Funding Total

-

Keywords

© Copyright 2026 RobotToday. All rights reserved powered by RobotToday.com