Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models

Benchmark Model Rank Results
robot-manipulation-on-liberoCoarse-to-Control–Success Rate: 97.9
robot-manipulation-on-libero-goalCoarse-to-Control–Success Rate: 97.8
robot-manipulation-on-libero-longCoarse-to-Control–Success Rate: 95.0
robot-manipulation-on-libero-objectCoarse-to-Control–Success Rate: 100.0
robot-manipulation-on-libero-spatialCoarse-to-Control–Success Rate: 98.8
robot-manipulation-on-simplerenv-widow-xCoarse-to-Control–Average: 0.833