Diffusion Policy
Columbia / Toyota Research Institute / collaborators
Visuomotor policy approach that represents robot behavior as a conditional denoising diffusion process.
- Code
- Not listed
- Weights
- Available
- Checkpoint
- Available
- License
- apache-2.0
Decision summary
Multi-modal action generation
Hardware, latency, dependencies, and checkpoint loading are not pipeline-tested.
Run a checkpoint loading smoke test before marking the model pipeline-tested.
Release history
Source-backed release timing for this canonical model record.
- Paper
Model series introduced
Release timing is anchored to the cited paper publication date.
Evidence profile
A visual read of adoption evidence. Scores describe catalog evidence readiness, not task performance.
Published evaluation
5/6
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
UsefulCheckpoint files were verified in the official repository metadata.
Training and loading reproducibility
UsefulConfiguration files are present; runtime loading is not tested.
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
observation
Action / robot state
actions · robot state · trajectory
Feedback / correction / failure
task success
Sim-real / embodiment metadata
robot state
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.
EgoWorld
Schema validation for LeRobot v3
provisional fit 83 · confidence 40
2 matched · 1 unknown
EgoStation GoPro Pick-and-Place
Pick-and-place benchmark fixtures
provisional fit 83 · confidence 40
2 matched · 1 unknown
Egocentric Adjust Bottle
LeRobot conversion tests
provisional fit 83 · confidence 40
2 matched · 1 unknown
More evidenceShowHide
Additional decision dimensions
Access and license are declared by the source.
75 · c85Required signal categories are structured; exact tensor and action interfaces still need verification.
76 · c70Hardware, latency, dependencies, and checkpoint loading are not pipeline-tested.
70 · c55Artifact facts
- release_year
- official_claim
- 2023
- curated official source year
- model.required_signals
- official_claim
- observation · actions · robot state · trajectory · task success
- Official model documentation and curated signal mapping
- model.hub
- official_claim
- declared
- curated model hub link
- model.weight_files
- metadata_verified
- repository: lerobot/diffusion_pusht · count: 1 · sample files: model.safetensors
- Hugging Face model metadata siblings
- model.checkpoint_files
- metadata_verified
- repository: lerobot/diffusion_pusht · count: 1 · sample files: model.safetensors
- Hugging Face model metadata checkpoint files
- model.config_files
- metadata_verified
- repository: lerobot/diffusion_pusht · count: 1 · sample files: config.json
- Hugging Face model metadata siblings
- license.weights
- metadata_verified
- apache-2.0
- Hugging Face model card license metadata
- license
- metadata_verified
- apache-2.0
- Hugging Face model card license metadata
- release_timing
- metadata_verified
- 2023-03-07T18:50:03.000Z
- arXiv 2303.04137 published
Additional references
OpenBot notes
- Important policy baseline for comparing whether a dataset needs a larger VLA/WAM or a stronger task policy.
- OpenBot failure mining should identify when diffusion policies fail from missing feedback or contact signals.