Does Diffusion Policy run on ALOHA (stationary)?
works with caveats
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | ev-oft-paper | |
| Real-robot inference | works | ev-oft-paper | |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Caveats
- Community-reported only; not documented upstream or maintainer-tested.
- Run by the OpenVLA-OFT authors as a from-scratch baseline using the DROID Diffusion Policy implementation (language-conditioned, modified for bimanual control), not LeRobot's diffusion policy.
- Struggled on the 'put X into pot' task, which had the largest training set.
Evidence 1 source; the cell shows the highest tier
- T3 · community reportedpapersource 2025-02-27verified 2026-10-24 (1d ago)
OpenVLA-OFT paper trains Diffusion Policy from scratch on four real tasks on the original ALOHA (two ViperX 300 S arms) and reports it matching or exceeding RDT-1B on clothes folding and scooping.
“Diffusion Policy demonstrates stronger capabilities, matching or exceeding RDT-1B’s reliability on the clothes folding and scooping tasks. However, it struggles with the “put X into pot” task which has a larger training dataset”
https://arxiv.org/abs/2502.19645
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/aloha/diffusion)<iframe src="https://caniembody.pages.dev/embed/aloha/diffusion" width="380" height="132" style="border:0" title="Diffusion Policy on ALOHA (stationary) | CanIEmbody"></iframe>History
- 2026-10-25 fecccd1 Initial release: robot policy compatibility matrix · Danilo Znamerovszkij