Reproduce “Safe Score Matching: Diffusion Policies with Hamilton-Jacobi Reachability for Online Safe Reinforcement Learning” 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

Boyang Li, Matthew Kim, Sylvia Lee Herbert

SSM trains online diffusion policies for control tasks where individual trajectories must satisfy safety constraints. A learned Hamilton–Jacobi critic routes guidance toward reward on viable states and toward reducing predicted worst-case violation during recovery. On Safety-Gymnasium Swimmer, SSM’s mean cost is 0.5 ± 0.4 versus 6.5 ± 2.4 for ALGD, but approximate critics and deterministic-dynamics assumptions preclude a safety guarantee.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.