← Leaderboard

π*0.6

VLA
Physical Intelligence · Nov 17, 2025 · built on π0.6

π0.6 trained with RECAP, an offline-then-on-robot RL recipe that conditions the policy on an advantage indicator and learns from demonstrations, autonomous rollouts and expert interventions. Results are self-reported on Physical Intelligence’s own tasks (laundry folding in real homes, box assembly, espresso making), where RECAP more than doubles throughput and roughly halves the failure rate on the hardest tasks.

Type
VLA
Parameters
—
Backbone
Gemma 3 4B + 860M flow-matching action expert
Weights
Closed
License
–
Commercial use
Unclear
Released
Nov 17, 2025
Robots
bimanual (two 6-DoF arms)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

0 of 13 benchmarks

No benchmark results recorded yet.

Not reported on 13 benchmarks
Wrong or missing number? Report an error