MMaDA-VLA: Large Diffusion Vision-Language-Action Model with Unified Multi-Modal Instruction and Generation

Benchmark Model Rank Results
zero-shot-generalization-on-calvinMMaDA-VLA (CALVIN ABC→D)#1Avg. sequence length: 4.78
robot-manipulation-on-libero-objectMMaDA-VLA#15Success Rate: 99.8
robot-manipulation-on-libero-spatialMMaDA-VLA#30Success Rate: 98.8
robot-manipulation-on-libero-goalMMaDA-VLA#35Success Rate: 98.0
robot-manipulation-on-libero-longMMaDA-VLA#48Success Rate: 95.2
robot-manipulation-on-liberoMMaDA-VLA#52Success Rate: 98.0