Does ACT run on Franka Emika Panda?
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | ev-levr-paper, ev-hub-tinkybala-act, ev-hub-ases-act | |
| Real-robot inference | works | LeVR: three tasks, 10 rollouts per task-policy pair, remote inference via LeRobot's gRPC policy server. | ev-levr-paper, ev-levr-paper-hw |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Caveats
- Community-reported only; not documented upstream or maintainer-tested.
- The real-robot evidence uses the LeFranX fork (Panda/FER + RobotEra XHand, 19-D action space), not an unmodified LeRobot install. LeFranX has no Franka Hand gripper interface.
Evidence 4 sources; the cell shows the highest tier
- T3 · community reportedpapersource 2025-09-17verified 2026-10-24 (1d ago)
LeVR paper: LeRobot ACT is fine-tuned on 100 LeFranX demos per task and rolled out on the robot (10 rollouts per task, 3 tasks).
“Using the LeRobot framework, we directly finetuned two state-of-the-art visuomotor policies, ACT and DP, on our LeFranX-collected dataset.”
https://arxiv.org/abs/2509.14349v1 - T3 · community reportedpapersource 2025-09-17verified 2026-10-24 (1d ago)
LeVR hardware: Franka FER with an XHand, three RealSense D435 cameras at 320x240/30 fps, inference on an RTX 6000 Ada server.
“Experiments were conducted using a Franka FER robot equipped with a RobotEra XHand dexterous hand.”
https://arxiv.org/abs/2509.14349v1 - T3 · community reportedhub-checkpointsource 2026-04-02verified 2026-10-24 (1d ago)
config.json type=act. train_config dataset TInkybala/PickCubeVLA0, whose meta/info.json robot_type=panda_ros_cartesian (148 episodes, panda_joint1-7 state, Cartesian-velocity actions). Auto-generated card, no rollout results. The Panda identity comes from the repo/robot_type naming.
https://huggingface.co/TInkybala/Real_Panda_PickCubeVLA0_act_20260402_143912 - T3 · community reportedhub-checkpointverified 2026-10-24 (1d ago)
config.json type=act. train_config dataset ases200q2/Real_Panda_CartesianVel_SpaceMouse, whose meta/info.json robot_type=panda_ros_cartesian (25 episodes, one camera).
https://huggingface.co/ases200q2/Real_Panda_CartesianVel_SpaceMouse_policy
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/franka-panda/act)<iframe src="https://caniembody.pages.dev/embed/franka-panda/act" width="380" height="132" style="border:0" title="ACT on Franka Emika Panda | CanIEmbody"></iframe>History
- 2026-10-25 fecccd1 Initial release: robot policy compatibility matrix · Danilo Znamerovszkij