PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations

Benchmark Model Rank Results
robot-manipulation-on-liberoPDRVLA–Success Rate: 99.3
robot-manipulation-on-libero-goalPDRVLA–Success Rate: 99.8
robot-manipulation-on-libero-longPDRVLA–Success Rate: 97.6
robot-manipulation-on-libero-objectPDRVLA–Success Rate: 99.8
robot-manipulation-on-libero-spatialPDRVLA–Success Rate: 100.0
robot-manipulation-on-robotwin-2-0-easy-50-tasksPDRVLA–Success Rate: 93.9
robot-manipulation-on-robotwin-2-0-hard-50-tasksPDRVLA–Success Rate: 93.3
robot-manipulation-on-simplerenv-widow-xPDRVLA–Average: 0.815