Does π0 run on Universal Robots UR5e?
works with caveats
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | ev-pi0-paper, ev-ur5e-gello-readme, ev-ur5e-lerobot-readme | |
| Real-robot inference | works | ev-pi0-paper, ev-ur5e-gello-readme, ev-ur5e-lerobot-readme | |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Caveats
- The upstream evidence is Physical Intelligence's own UR5e evaluation; openpi ships no UR5e fine-tuned checkpoint, and its examples/ur5 is an outline for writing UR5 input/output transforms rather than a tested recipe.
- Public end-to-end UR5e deployments are community-reported only (an openpi fork driven through a LeRobot plugin, and a LeRobot 0.5.1 fork).
Evidence 4 sources; the cell shows the highest tier
- T1 · upstream documentedpapersource 2024-10-31verified 2026-10-24 (1d ago)
The π0 paper lists UR5e (one arm, two cameras, 7-D action space) among its platforms and fine-tunes and evaluates π0 on a UR5e bowl-stacking task.
“UR5e stack bowls. This task requires stacking bowls, with four bowls of different sizes.”
https://arxiv.org/abs/2410.24164 - T1 · upstream documentedsource-codesource 2026-08-24verified 2026-10-24 (1d ago)
openpi's UR5 example only outlines how to write UR5 input/output transforms for fine-tuning; it is not a released checkpoint or tested recipe.
“Below we provide an outline of how to implement the key components mentioned in the "Finetune on your data" section of the [README](../README.md) for finetuning on UR5 datasets.”
https://github.com/Physical-Intelligence/openpi/blob/215abfb217dbac7d5f1273282331b9b1866c0479/examples/ur5/README.md - T3 · community reportedsource-codesource 2026-02-16verified 2026-10-24 (1d ago)
Third-party UR5e LeRobot plugin (Robotiq gripper, RTDE) records datasets with LeRobot and runs π0 checkpoints served by an openpi fork (openpi-ur5e, config pi0_ur5e) on the real UR5e, with an evaluation script.
“**Remote policy inference** — deploy OpenPI (π0) policies via WebSocket from a GPU server”
https://github.com/F-Fer/lerobot_ur5e_gello/blob/3f3532cfc58d81ba12c1582f9c6087d12488269d/README.md - T3 · community reportedsource-codesource 2026-07-02verified 2026-10-24 (1d ago)lerobot 0.5.1
LeRobot route: the UR5e-LeRobot fork (lerobot 0.5.1) trains --policy.type=pi0 from lerobot/pi0_base and evaluates it on single-arm and bimanual UR5e.
“We have validated and tuned hyperparameters on this hardware stack for five policies: `ACT`, `Diffusion Policy`, `VQ-BeT`, `MultiTaskDiT`, `pi0`, and `pi05`.”
https://github.com/yechen056/UR5e-LeRobot/blob/72d41bdae2606f464d3fd0da24cd6a20241ab244/README.md
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/ur5e/pi0)<iframe src="https://caniembody.pages.dev/embed/ur5e/pi0" width="380" height="132" style="border:0" title="π0 on Universal Robots UR5e | CanIEmbody"></iframe>History
- 2026-10-25 fecccd1 Initial release: robot policy compatibility matrix · Danilo Znamerovszkij