Facet-0: A Robotic Foundation Model for Contact-Rich Precise Manipulation
Facet-0 reports 82% mean success on five sub-millimeter computer-assembly tasks, versus 15% for the strongest baseline.
The arXiv paper describes a robotic foundation model built for contact-heavy assembly, where force feedback matters as much as visual progress. It predicts action chunks together with the wrist-wrench profile those actions should produce, then uses rollout data to score contact outcomes. The system was trained on ManuFacet-1K, a 1,000-hour force-synchronized dataset across three robot embodiments and multiple manufacturing cells. The authors report 0.5 mm placement accuracy and 50 ms command latency in the bounded task-adapted version. ArXiv · AI/CL/LG's note
The arXiv paper describes a robotic foundation model built for contact-heavy assembly, where force feedback matters as much as visual progress. It predicts action chunks together with the wrist-wrench profile those actions should produce, then uses rollout data to score contact outcomes. The system was trained on ManuFacet-1K, a 1,000-hour force-synchronized dataset across three robot embodiments and multiple manufacturing cells. The authors report 0.5 mm placement accuracy and 50 ms command latency in the bounded task-adapted version. ArXiv · AI/CL/LG's note
score 6