Accelerating Vision-Language-Action Model Integrated with Action Chunking via Parallel Decoding

Benchmark Model Rank Results
robot-manipulation-on-calvin-abcd-to-dPD-VLA (CALVIN ABCD→D)–Average Length: 3.54
robot-manipulation-on-liberoPD-VLA–Success Rate: 94.7
robot-manipulation-on-libero-goalPD-VLA–Success Rate: 94.9
robot-manipulation-on-libero-longPD-VLA–Success Rate: 91.7
robot-manipulation-on-libero-objectPD-VLA–Success Rate: 96.7
robot-manipulation-on-libero-spatialPD-VLA–Success Rate: 95.5