Reproduce “LEASH: Stay Safe in Agentic Robotics” 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
Qinfeng Zhu, Weiguang Zhao, Ke Sun, Xihao Guo, Yunxi Jiang, Rui Zhang, Lei Fan, Kaizhu Huang
LEASH adds model-agnostic safety checks between planner and robot, reviewing goals, proposed actions, and selected operations in visual context. For GPT-6 Astra in RoboRisk-S simulation, it reduced task-averaged Harm Rate from 11.23% to 0.00% and raised Safe Success Rate from 78.99% to 80.80%. This result covers one planner in simulation; it does not establish that LEASH eliminates harm.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.