Reproduce “Coachable agents for interactive gameplay” 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
Roberto Capobianco, Harm van Seijen, Nolan D. Bard, Neil Burch, Fatima Davelouis, Josh Davidson, Alisa Devlic, Yunshu Du and 33 more
The paper presents a single control policy that changes its behavior in response to structured style settings, while keeping the task fixed. During training, task rewards are combined with style-dependent rewards; at execution, changing the setting selects a behavior without retraining. In Horizon Forbidden West combat, default and melee styles won 99.8% and 83.4% of battles in selected 1-vs-1 scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.