OM-1
Reward AI
OM-1 (Omnibody Model 1) is Reward AI's general-purpose robot policy learned only from human demonstrations captured with the wearable Omnibody Hand, with no teleoperation or on-robot data, and deployed across robot arms…
Source notes
OM-1 (Omnibody Model 1) is Reward AI's general-purpose robot policy learned only from human demonstrations captured with the wearable Omnibody Hand, with no teleoperation or on-robot data, and deployed across robot arms and humanoids at human speed.
- Code
- Unverified
- Weights
- Unverified
- Checkpoint
- Unverified
- License
- Unverified
Decision summary
Learning contact-rich manipulation from robot-free human data
Workflows that require publicly downloadable weights or code
Verify repository and artifact licenses.
Release history
Source-backed release timing for this canonical model record.
- Official video
Model release
Release timing is recorded from official model metadata.
Evidence profile
A visual read of adoption evidence. Scores describe catalog evidence readiness, not task performance.
Published evaluation
1/6
dimensions scored
Unknown evidence remains visible and is never treated as a zero.
Why these scores
The strongest decision reasons behind the evidence profile.
Artifact availability
UnknownCode and weights are not verified.
Training and loading reproducibility
UnknownNo verified loading configuration is available.
Evaluation evidence
UnknownNo structured evaluation evidence has been verified.
Data requirements
Declared loop-data needs, missing evidence, and the next checks that matter.
Observation / ego video
Images
Action / robot state
Hand pose trajectories
Sim-real / embodiment metadata
Descriptive model record
Critical gaps
Core categories are represented. Interface alignment and data quality still require verification.
Linked datasets
Signal-level links only. Verify runtime interfaces before use.
No related catalog datasets are linked to this model yet.
More evidenceShowHide
Additional decision dimensions
License or access terms are not verified.
— · c20Required signal categories are structured; exact tensor and action interfaces still need verification.
76 · c70Hardware, latency, dependencies, and checkpoint loading are not pipeline-tested.
— · c10Artifact facts
- catalog.curation
- openbot_inference
- tier: editorial_focus · collection: OpenBot important model releases · policy: human_curated_priority
- OpenBot editorial release selection
- model.type
- official_claim
- General-purpose robot policy
- Official release section Omnibody Model 1
- model.reported_training_data
- official_claim
- source: robot-free human demonstrations via Omnibody Hand · teleoperation: not used · on robot data: not used · new task data: less than 30 minutes
- Official release sections Omnibody Model 1 and Conclusion
- model.reported_interface
- official_claim
- inputs: images · tactile signals · inter-finger proximity · hand pose trajectories · outputs: motion direction · speed · force · key-event timing · control layer: reinforcement learning in simulation
- Official release sections Omnibody Model 1 and Control Any Body
- model.required_signals
- openbot_inference
- Images · Tactile signals · Inter-finger proximity · Hand pose trajectories · Force
- OpenBot signal mapping from the official input and action description
- release_timing
- metadata_verified
- 2026-09-14T00:00:00.000Z
- Existing published Catalog release-date evidence
Additional references
OpenBot notes
- Task-learning speed, tracking accuracy, and cross-embodiment results are source-reported claims, not OpenBot pipeline tests.
- Training data depends on Reward AI's Omnibody Hand and One Data Interface capture hardware.
- The official release does not list a public code, weight, checkpoint, or license artifact.