Reproduce “Success Leaves Detours: Learning Executable Walkthroughs for Long-Horizon Agents” 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
Kaijie Chen, Chenyu Fang, Liang Yan, Bo Li, Bo Zhang, Peng Ye
Trace converts sparse-reward attempts into reusable procedures by tracing progress backward and estimating which state facts later actions depend on. It stores step conditions, actions, expected effects, and entry, completion, and failure checks. Across WebShop, J-TTL, and ScienceWorld, it improves reported averages over strongest baselines by 30.0% in AUC and 40.5% in Final-3; prerequisites remain observational associations, not proven necessities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.