Facet-0: a contact-aware foundation model for precise assembly
Loading video
Loading videoFacet-0 predicts and values the contact consequences of actions, combining multimodal representation learning and RL post-training around joint action-wrench generation. On five sub-millimeter computer assembly tasks it reaches 82% mean success versus 15% for the strongest baseline, with 0.5 mm placement accuracy and 50 ms command latency.
接触操作精密装配亚毫米操作机器人基础模型VLA力控力同步数据集ManuFacet-1KFacet-0PINE Labcontact-rich manipulationprecise assemblyrobotic foundation modelforce controlrobotsdigestTwitter
Category: industrial
Author: @robotsdigest
Date: 2026-09-02T00:00:00
Duration: 19.233s
Reference: https://arxiv.org/abs/2609.01596
