Reproduce “Active Reward Learning from Online Preferences” 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

Vivek Myers, Erdem Bıyık, Dorsa Sadigh

The authors present an online robot-guidance method that represents unknown preferences as a posterior over reward functions. At each state, it compares candidate action pairs, estimates each query’s expected value of information, and asks only when the best value exceeds a threshold. Across simulations and a 22-person driving study, it reduced queries and speed non-adherence versus uncertainty-triggered querying, without establishing faster execution or general real-world human deployment.

Summary by alphaXiv.

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