Reproduce “Teaching Vision-Language-Action Models to Stop Safely” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zheyuan Liu, Chuhan Wang, Jinduo Guo, Yue Huang, Xiangchi Yuan, Ningshan Ma, Meng Jiang
SAFI teaches policies to handle mid-task takeover during prohibited tasks by supervising a stationary arm that maintains its existing gripper command across task states. It also trains on permitted tasks. In simulated LIBERO Goal, Spatial, and Object evaluations, target success fell from 99.7% to 1.3%, with zero recorded incidents or action-saturation episodes; results cover selected tasks where holding is appropriate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.