Facet-0: A Robotic Foundation Model for Contact-Rich Precise Manipulation
Facet-0 unifies multimodal representation learning and reinforcement learning post-training around a joint action-wrench proposal, which reaches 82% mean success on five sub-millimeter computer-assembly tasks, compared with 15% for the strongest baseline.
Hao-Yuan Deng, Hai-Chao Liu, Wen-Kai Guo et al.
· 1 citation