Reproduce “Learning Safe Flow Matching Robot Policies from Binary Feasibility Oracles” 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
Jianming Ma, Rongjun Jin, Yizhi Chen, Yue Gao
The authors introduce an oracle-guided safeguard for flow-matching robot policies that uses binary feasibility checks to shorten unsafe updates and estimate local boundary information for training. Applied during both training and inference, it achieves 100% step safety and 11 successful episodes out of 30 in a simulated bottle task, versus 78.25% and 13 for the baseline, at higher inference time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.