← Leaderboard

SpatialVLA

VLA
Shanghai AI Lab · Jan 27, 2025

VLA from Shanghai AI Lab with ShanghaiTech, TeleAI and several universities. It adds Ego3D position encoding and adaptive spatial action grids to PaliGemma 2 and is pretrained on 1.1M Open X-Embodiment and RH20T episodes. Its LIBERO and SimplerEnv numbers are self-reported.

Type
VLA
Parameters
3.5B
Backbone
PaliGemma 2 (3B)
Weights
Open
License
MIT
Commercial use
Allowed
Released
Jan 27, 2025
Robots
Google Robot (sim), WidowX, Franka Panda (sim), Franka
SO-101
–
LeRobot
–
Inputs
RGB · L
VRAM
8.5GB

Benchmark results

2 of 13 benchmarks
LIBEROSaturatedSimulation
#65 of 7078.1%, self-reported by the authors, not a leaderboard entry
SimplerEnv WidowXSaturatedSimulation
#24 of 3042.7%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Wrong or missing number? Report an error