Native Action-Prior Learning from Videos for World Action Models

Benchmark Model Rank Results
robot-manipulation-on-liberoNAVA-WAM–Success Rate: 99.0
robot-manipulation-on-libero-plus-zero-shot-transferNAVA-WAM–Success Rate: 83.5
robot-manipulation-on-robotwin-2-0-easy-50-tasksNAVA-WAM–Success Rate: 88.5
robot-manipulation-on-robotwin-2-0-hard-50-tasksNAVA-WAM–Success Rate: 73.6