RewardAI Launches OM-1: Robot Foundation Model Zero-Shot Trained on Pure Human Data

On September 15, RewardAI officially launched the Omnibody technology stack and its first robot foundation model OM-1 (core team includes Chen Wang, Zhu Yifeng, Zipeng Fu and others). OM-1's core principle is "One Model, One Data Interface, Any Body": a single model zero-shot controls desktop arms, industrial arms and even humanoids (paired with 7-DoF Omnibody Hands) without per-robot fine-tuning. The announcement was amplified by Zipeng Fu, Yifeng Zhu and other researchers, with OpenAI researchers also resharing praise for motion fluency.

Confirmed

• Training learns directly from human manipulation data: no teleoperation data, no robot-collected data, no pre/post-training split or robot-specific fine-tuning; skills learned from human motion transfer directly across robots
• Claims near-human dexterity and efficiency, multi-robot collaboration; demo videos are stated to be real-time with no speedup
• Emergent behaviors: dual arms compensate for each other's mistakes; the model decides when to retry, adapt to adversarial perturbations, or stop under drastic environment changes
• Efficiency: long-horizon tasks compressed—four-hand phone packing, cocktail making and laundry folding in under 30 seconds; team stresses "efficiency is ROI, and ROI determines whether robots can be deployed"
• Data collection uses 30 minutes of human demonstration wearing the Omnibody Hand rather than teleoperation
• Some reposts (kenziyuliu, zipengfu) attribute the release to Physical Intelligence, inconsistent with most posts crediting RewardAI; the RewardAI official account is taken as authoritative

Unconfirmed

• Zero-shot cross-embodiment generalization and near-human dexterity rest mainly on official demo videos and team claims, pending independent third-party replication or evaluation

Why it matters

• If zero-shot cross-embodiment transfer holds amid monthly hardware iterations, it could sharply cut retraining/fine-tuning costs per robot platform
• Training on human (not teleoperation) data plus emergent dual-arm error correction points toward general-purpose robot assistants
• The "efficiency is ROI" narrative ties research capability directly to commercial deployment, a key argument in the robot foundation model race

2026-09-15 ~ 2026-09-16 · 51 related posts

• Reward AI debuts OM-1 robot foundation model, zero-shot generalizing to any robot — yifengzhu_ut · 2026-09-15
• RewardAI's OM-1 zero-shot generalizes across arms and humanoids — zipengfu · 2026-09-15
• Reward AI's OM-1 does quadmanual packaging, bartending, laundry folding in under 30 seconds, real-time — zipengfu · 2026-09-15
• Reward AI launches OM-1: one robot policy zero-shot across arms, industrial arms and humanoids — zipengfu · 2026-09-15
• Reward AI's OM-1 shows emergent behaviors: arms compensating for each other's mistakes — zipengfu · 2026-09-15
• OM-1 robot foundation model generalizes zero-shot across arms and humanoids from human data — zipengfu · 2026-09-15
• Physical Intelligence unveils OM-1, a robot foundation model trained purely on human data — zipengfu · 2026-09-15
• Physical Intelligence launches OM-1 robot foundation model with zero-shot generalization — zipengfu · 2026-09-15
• Reward AI debuts Omnibody Hand & OMI-1, pushing hardware-algorithm co-design for cross-embodiment dexterity — zipengfu · 2026-09-15
• OM-1 robot foundation model learns from human data, generalizes zero-shot across robots — zipengfu · 2026-09-15
• Zipeng Fu's RewardAI team draws praise for OM-1 robot foundation model — zipengfu · 2026-09-15
• RewardAI launches OM-1 foundation model trained on human movement data — lukas_m_ziegler · 2026-09-15
• [source] RewardAI's OM-1 Robot Foundation Model Zero-Shot Generalizes to Arms and Humanoids — zipengfu · 2026-09-15
• Reward's OM-1 learns new manipulation tasks from under 30 minutes of human demos — chris_j_paxton · 2026-09-15

36 near-duplicate retellings: zipengfu · zipengfu · zipengfu · zipengfu · zipengfu · zipengfu · zipengfu · kenziyuliu · zipengfu · yifengzhu_ut · yifengzhu_ut · zipengfu · zipengfu · lukas_m_ziegler · yifengzhu_ut · SimplyAnnisa · yifengzhu_ut · yifengzhu_ut · bigaiguy · yifengzhu_ut · zipengfu · zipengfu · yifengzhu_ut · yifengzhu_ut · zipengfu · yifengzhu_ut · yifengzhu_ut · yifengzhu_ut · yifengzhu_ut · yifengzhu_ut · zipengfu · zipengfu · yifengzhu_ut · yifengzhu_ut · yifengzhu_ut · yifengzhu_ut