← Leaderboard

villa-X

VLA
Microsoft Research Asia · Jul 31, 2025

VLA from Microsoft Research with Tsinghua and others that learns latent actions from video and uses them to guide a PaliGemma-based policy. Only the latent action model is released; the policy weights are not.

Type
VLA
Parameters
—
Backbone
PaliGemma 3B + latent action model
Weights
Open
License
MIT
Commercial use
Allowed
Released
Jul 31, 2025
Robots
Realman RM75, XArm with XHand, WidowX (sim), Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

2 of 13 benchmarks
LIBEROSaturatedSimulation
#56 of 7090.1%, self-reported by the authors, not a leaderboard entry
SimplerEnv WidowXSaturatedSimulation
#16 of 3062.5%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Wrong or missing number? Report an error