VLM4VLA: Revisiting Vision-Language-Models in Vision-Language-Action Models

Benchmark Model Rank Results
robot-manipulation-on-calvinVLM4VLA (Qwen3VL-2B, CALVIN ABC-D)–avg. sequence length (D to D): 4.142
robot-manipulation-on-libero-longVLM4VLA (Qwen3VL-2B, LIBERO-Long)–Success Rate: 55.8
robot-manipulation-on-simplerenv-widow-xVLM4VLA (KosMos-2, SimplerEnv Bridge/WidowX)–Average: 0.604