Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models

Benchmark Model Rank Results
robot-manipulation-on-robotwin-2-0-easy-50-tasksVLAct (PI, data scaling)#9Success Rate: 93
robot-manipulation-on-robotwin-2-0-hard-50-tasksVLAct (OFT, data scaling)#17Success Rate: 90.8
robot-manipulation-on-libero-plus-zero-shot-transferVLAct (PI)#18Success Rate: 82.6