G^3VLA: Geometric Inductive Bias for Vision-Language-Action Models

Benchmark Model Rank Results
robot-manipulation-on-liberoG^3VLA (GT)–Success Rate: 88.1
robot-manipulation-on-libero-goalG^3VLA (GT)–Success Rate: 88.4
robot-manipulation-on-libero-objectG^3VLA (GT)–Success Rate: 94.4
robot-manipulation-on-libero-spatialG^3VLA (GT)–Success Rate: 89.2