Reproduce “Successor Feature Landmarks for Long-Horizon Goal-Conditioned 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
Christopher Hoang, Sungryull Sohn, Jongwook Choi, Wilka Carvalho, Honglak Lee
SFL uses shared successor features (SF) to localize states, build a landmark graph, and derive actions toward nearby landmarks. It explores graph frontiers and plans routes through shorter segments. In fixed-start ViZDoom Test-1, Hardest, SFL reached goals in 50% of evaluations, versus 1% for MSS. These results cover tested simulated settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.