Reproduce “Compositional Diffusion Planning with Chunk Selection and Reachability Measure” 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
Chanyoung Park, Youngjoong Kim, Jaesik Park
SCD composes long-horizon maze plans by selecting compatible segments from generated trajectory chunks using a learned reachability score, then generating bridges and refining the assembled route. Its classifier uses generated negatives and density reweighting. On HumanoidMaze-Giant, it reports 83±3% success versus 67±4% for CompDiffuser and 62±3% for RCD; 76% of plans are geometrically feasible, not guaranteed executable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.