# OM-1: A Robot Policy Trained Exclusively on Human Demonstrations

Digest AI · Enterprise & Industry · published 2026-09-14T21:08:17Z

Canonical: https://digestai.news/story/om-1-a-robot-policy-trained-exclusively-on-human-demonstrations

## Summary

Reward AI, a robotics startup known for its DexCap and HumanPlus projects, has unveiled OM-1. This general-purpose manipulation policy learns from human wearers equipped with a sensorized glove, capable of operating industrial arms and humanoid robots at normal speed. Key findings include the absence of teleoperation or on-robot data in training. The system's unique approach is termed 'One Model, One Data Interface, Any Body,' emphasizing that capturing demonstrations can train any robot embodiment. OM-1’s success hinges on a wearable glove with seven degrees of freedom (DOF), capable of capturing complex hand movements and force sensations. This design allows for high-speed interactions without needing to calibrate each user individually. The system's core innovation lies in its ability to learn from human demonstrations, bypassing the need for traditional robot data or teleoperation methods. Reward AI attributes OM-1’s effectiveness to a novel architecture that integrates both the manipulation policy and control layer into a single pipeline, enabling real-time learning without intermediate stages. This approach not only accelerates training but also ensures robust performance across different robotic platforms. The company claims that under 30 minutes of human data is sufficient for learning new tasks, setting a new standard in robot manipulation research.

## Key points

- OM-1 learns from human demonstrations using a sensorized glove with seven DOF
- No teleoperation or on-robot data used during training
- Capable of operating various robotic platforms including arms and legged robots

## Why it matters

This development marks a significant step forward in robot manipulation, particularly for industrial applications. By eliminating the need for traditional robot-centric datasets and teleoperation methods, OM-1 opens up new possibilities for deploying advanced robotics without requiring extensive infrastructure or specialized training.

## Sources

1. [Reward AI Releases OM-1: A Robot Policy Trained on Human Demonstrations Only, With No Teleoperation or On-Robot Data](https://marktechpost.com/2026/09/14/reward-ai-releases-om-1-a-robot-policy-trained-on-human-demonstrations-only-with-no-teleoperation-or-on-robot-data) (MarkTechPost, 2026-09-14)

## Cite

Digest AI, "OM-1: A Robot Policy Trained Exclusively on Human Demonstrations", 14 September 2026, https://digestai.news/story/om-1-a-robot-policy-trained-exclusively-on-human-demonstrations

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/om-1-a-robot-policy-trained-exclusively-on-human-demonstrations.json