GeoVLA: Empowering 3D Representations in Vision-Language-Action Models

Benchmark Model Rank Results
robot-manipulation-on-liberoGeoVLA–Success Rate: 97.7
robot-manipulation-on-libero-90GeoVLA–Success Rate: 97.7
robot-manipulation-on-libero-goalGeoVLA–Success Rate: 96.6
robot-manipulation-on-libero-longGeoVLA–Success Rate: 96.6
robot-manipulation-on-libero-objectGeoVLA–Success Rate: 99.0
robot-manipulation-on-libero-spatialGeoVLA–Success Rate: 98.4