Does LingBot-VA run on SO-101?
partial
Scope: Training on SO-101 data is evidenced (an upstream contributor reproduced post-training on a user's SO-101 dataset); on-robot inference is reported only with low success and jitter in an open, unresolved issue.
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | Upstream contributor converted the 83-episode SO-101 dataset and trained it with the official post-training code (8 GPUs, reasonable visualizations after ~200 steps). | ev-gh-lingbotva-29, ev-dataset-pick-n-place-sq |
| Real-robot inference | no verified evidence | — | |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Caveats
- Community-reported only; not documented upstream or maintainer-tested.
- The reporter's SO-101 deployments (open-loop and closed-loop with KV-cache updates) produced random or back-and-forth motion and only one, apparently accidental, success. A later commenter attributes the jitter to an action-reshaping bug in the reporter's client script; no SO-101 success is confirmed in the thread.
Evidence 2 sources; the cell shows the highest tier
- T3 · community reportedgithub-issuesource 2026-02-26verified 2026-10-25 (today)
Issue 'Finetuning on custom pick and place dataset on SO-101 arm' (open): user fine-tuned robbyant/lingbot-va-base on an SO-101 dataset and deployed it with poor results; contributor lilin-hitcrt reproduced training on the converted dataset and shared a debug config (wan_va/configs/va_demo_train_cfg.py).
“I've converted your dataset to the proper format and tested it with our training code. Following the provided data format and default parameters, the model trains successfully on 8 GPUs and shows good visualization results after ~200 steps.”
https://github.com/Robbyant/lingbot-va/issues/29 - T3 · community reportedcommunity-postverified 2026-10-25 (today)
The dataset used in the issue: meta/info.json robot_type=so101_follower, LeRobot v3.0 format, 83 episodes, top + wrist cameras.
https://huggingface.co/datasets/AkashKarnatak/pick-n-place-sq
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/so101/lingbot-va)<iframe src="https://caniembody.pages.dev/embed/so101/lingbot-va" width="380" height="132" style="border:0" title="LingBot-VA on SO-101 | CanIEmbody"></iframe>History
- 2026-10-25 fecccd1 Initial release: robot policy compatibility matrix · Danilo Znamerovszkij