NORA-1.5: A Vision-Language-Action Model Trained using World Model- and Action-based Preference Rewards

Benchmark Model Rank Results
robot-manipulation-on-simplerenv-google-robotNORA-1.5 (DPO, Visual Matching)#2Success Rate: 82.8
robot-manipulation-on-libero-spatialNORA-1.5 (DPO) (LIBERO-Spatial)#54Success Rate: 98.0
robot-manipulation-on-libero-goalNORA-1.5 (DPO) (LIBERO-Goal)#80Success Rate: 95.4
robot-manipulation-on-libero-longNORA-1.5 (DPO) (LIBERO-Long)#81Success Rate: 90.5
robot-manipulation-on-libero-objectNORA-1.5 (DPO) (LIBERO-Object)#90Success Rate: 96.0
robot-manipulation-on-liberoNORA-1.5 (DPO)#109Success Rate: 95.0