Reproduce “A Framework for Transforming Specifications in 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
Rajeev Alur, Suguman Bansal, Osbert Bastani, Kishor Jothimurugan
The authors define a framework for testing whether logical control tasks in unknown environments can be transformed using samples while preserving policy guarantees. They show how to simulate such reductions without access to transition probabilities and analyze reward-based representations. Their main result is that no PAC-MDP algorithm covers the class of safety specifications, demonstrated with one fixed safety formula.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.