Reproduce “Learning Rewards in Strategic Games” 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
Antoine Bergerault, Cyrille Kone, Giorgia Ramponi
Active IRL learns follower rewards through deliberately chosen interactions in Stackelberg games with known linear features and an entropy-regularized follower. It selects informative leader policies and evaluates learning by downstream leader payoff, not reward error. In a Lobéké transfer task after 1,600 poacher actions, mean SubOpt was 2.0×10⁻⁴ versus 3.6×10⁻³ for Fixed-expert MLE; this is illustrative, not a general transfer guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.