Does Diffusion Policy run on SO-100?
works with caveats
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | ev-hub-hcltech-diffusion-so100, ev-hub-neilkim-diffusion-so100, ev-van-i-so100-stand | |
| Real-robot inference | works with caveats | 9/10 at trained positions, 0/2 outside the demonstrated area; ~16 s to first arm movement. | ev-van-i-so100-stand |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Caveats
- Community-reported only; not documented upstream or maintainer-tested.
- The physical arms are SO-100, but the run used --robot.type=so101_follower (LeLab uses that type for both arm versions); reproduce with the same driver type.
- One community setup (three 640x480 cameras, one pick-and-place task): 9/10 at trained start positions, 2/2 between them, 0/2 outside the marked area.
Evidence 3 sources; the cell shows the highest tier
- T3 · community reportedmodel-cardsource 2026-10-04verified 2026-10-25 (today)lerobot 0.6.0
Write-up of a self-built SO-100 leader/follower stand (LeRobot 0.6.0 via LeLab, RTX 3090): Diffusion Policy trained from scratch for 36,000 steps and tested on the physical arm, 9/10 at trained positions and 0/2 outside the marked area. States so101_follower is the LeLab driver type for both arm versions.
“SO-100 leader (moved by hand) + SO-100 follower (does the task), built from parts”
https://huggingface.co/van-i/so100-imitation-learning-stand/blob/c68846b72dea1e43f63a485684dc1a7d3e3d2ade/README.md - T3 · community reportedhub-checkpointsource 2025-08-12verified 2026-10-24 (1d ago)
config.json type=diffusion; train_config.json dataset.repo_id=targabor/so100_ducktape_3cam, whose meta/info.json robot_type=so100_follower (50 episodes).
https://huggingface.co/hcltech-robotics/diffusion_so100_ducktape_3cam - T3 · community reportedhub-checkpointsource 2025-07-02verified 2026-10-24 (1d ago)
config.json type=diffusion; train_config.json dataset.repo_id=NeilKim/so100-teleop-record-2, whose meta/info.json robot_type=so100_follower (61 episodes).
https://huggingface.co/NeilKim/diffusion-so100-policy-2
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/so100/diffusion)<iframe src="https://caniembody.pages.dev/embed/so100/diffusion" width="380" height="132" style="border:0" title="Diffusion Policy on SO-100 | CanIEmbody"></iframe>