← Back to Open Data
verifiedsource · paperenv · pi0 paper Franka eval (pick / pour / fold subset)

Franka Research 3×pi0 base

Franka Robotics · LeRobot · lerobot

Evidence

Success rate
85.0%
Episodes tested
300
Updated
2026-04-20

https://arxiv.org/abs/2410.24164

Notes

Canonical exemplar. pi0 is Physical Intelligence's flow-matching VLA; Franka Research 3 is its primary research-arm target. Highest success rate on this list and the cleanest paper-to-prod path — the LeRobot port replays the paper's evaluation slice directly.

Gaps noted

  • · Reported numbers cover the ALOHA-style observation space (3 cameras + proprio). Single-camera deployment lands ~10 points lower in the paper's ablation.

Robot · Franka Research 3

Franka Research 3, 7-DoF torque-controlled research arm. Successor to Franka Emika Panda. De facto manipulation-research standard.

View robot profile →

Policy · pi0 base

Physical Intelligence pi0 vision-language-action base policy. 3.5B params, trained on cross-embodiment manipulation data. Strong baseline for dexterous manipulation including fold-laundry.

View policy profile →