Does HIL-SERL run on SO-100?
works with caveats
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | Online RL on the physical robot (actor/learner). | ev-lerobot-hilserl-learner |
| Real-robot inference | works | ev-lerobot-hilserl-learner | |
| Simulation inference | no verified evidence | — | |
| Teleop + record | works with caveats | Gamepad/keyboard documented; SO leader as intervention device reported broken (#2952). | ev-lerobot-hilserl-doc, ev-gh-lerobot-2952 |
Caveats
- The guide's documented intervention path is a gamepad or keyboard in end-effector space (needs a URDF); using an SO leader arm as the intervention device is reported broken in open issue #2952, not yet confirmed or fixed by maintainers.
- src/lerobot/configs/train_config_hilserl_so100.json is referenced by the guide but is not in the v0.6.1 source tree; the example configs are hosted in the aractingi/lerobot-example-config-files dataset, and its env_config_so100.json still names the removed so100_follower_end_effector robot type.
Evidence 3 sources; the cell shows the highest tier
- T1 · upstream documentedofficial-docssource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
LeRobot v0.6.1 HIL-SERL real-robot guide: find EE bounds and collect demos on so100_follower/so100_leader, then run the actor/learner with train_config_hilserl_so100.json.
“lerobot-find-joint-limits \ --robot.type=so100_follower \”
https://github.com/huggingface/lerobot/blob/v0.6.1/docs/source/hilserl.mdx - T1 · upstream documentedofficial-docssource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
LeRobot v0.6.1 HIL-SERL guide launches the learner and actor with an SO100 HIL-SERL training config (on-robot RL).
“python -m lerobot.rl.learner --config_path src/lerobot/configs/train_config_hilserl_so100.json”
https://github.com/huggingface/lerobot/blob/v0.6.1/docs/source/hilserl.mdx - T3 · community reportedgithub-issuesource 2026-02-10verified 2026-10-24 (1d ago)lerobot 0.4.4
Open issue (LeRobot 0.4.4, labelled bug): the SO leader arm cannot be used as the HIL-SERL intervention device; a 2026-08-02 comment against main reports control_mode is never read by the RL pipeline.
“TypeError: Teleoperator SOLeader must implement get_teleop_events() method. Compatible teleoperators: GamepadTeleop, KeyboardEndEffectorTeleop”
https://github.com/huggingface/lerobot/issues/2952
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/so100/hil-serl)<iframe src="https://caniembody.pages.dev/embed/so100/hil-serl" width="380" height="132" style="border:0" title="HIL-SERL on SO-100 | CanIEmbody"></iframe>History
- 2026-10-25 fecccd1 Initial release: robot policy compatibility matrix · Danilo Znamerovszkij