Spa-Bench/spa-bench-model-pi-0-5-step-076596

pi0.5 — Spa-Bench Step 76,596

This checkpoint is released as an anonymous supplementary artifact for a paper under double-blind review. It was evaluated on a physical SO-101 robot in Spa-Bench.

Model details

Field	Value
Checkpoint	End of epoch 12; step 76,596
Inputs	Middle RGB, wrist RGB, six joint positions, and a text instruction
Outputs	Six absolute joint-position targets
Action horizon	50
Adaptation	Vision-language backbone and action expert updated; no PEFT
Optimizer	AdamW; peak learning rate 2.5e-5; weight decay 0.01

Training-data documentation: spa-bench-training-teleoperation-1200.

Physical evaluation

82/120 familiar trials, 122/300 withheld-composition trials, 54/120 matched withheld trials, and 100/120 matched direct-manipulation controls. These are physical rollout results, not simulation metrics.

Rollouts: spa-bench-eval-rollouts-pi-0-5-full.

Limitations and safety

The rollout repository contains 920 recordings. Episodes 0–899 are scored; episodes 900–919 are additional Relative Size recordings excluded from reported results. Robot policies can move hardware unexpectedly. Use conservative limits, an accessible emergency stop, a clear workspace, and direct supervision. Do not deploy unattended or in safety-critical settings.

Double-blind release note

Author, institution, source-repository, and archival citation details are intentionally omitted during review. They will be restored in the archival release.