← Leaderboard

F1-VLA

VLA
Shanghai AI Lab · Sep 8, 2025 · built on π0

VLA that generates a foresight image of the next visual state and plans actions toward it, initialized from π0. Real-robot results are self-reported (82.2% over 9 Genie tasks vs 65.2% for π0).

Type
VLA
Parameters
4.2B
Backbone
π0 experts + foresight image-generation expert
Weights
Open
License
CC BY-NC-SA 4.0
Commercial use
Not allowed
Released
Sep 8, 2025
Robots
AgiBot Genie-1, Franka, ARX LIFT II, WidowX (sim), Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

2 of 13 benchmarks
LIBEROSaturatedSimulation
#49 of 7095.7%, self-reported by the authors, not a leaderboard entry
SimplerEnv WidowXSaturatedSimulation
#18 of 3059.4%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Wrong or missing number? Report an error