Reproduce “Marginal-Preserving Coupled Diffusion Composition for Long-Horizon Planning” 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

Hengrui Zhang, Qiyu Wang, Yuhu Cheng, Brooks Paige, Xuesong Wang

MCDC changes initial noise dependence among overlapping trajectory chunks while preserving each chunk’s standard-Gaussian marginal and leaving the pretrained planner and stitching procedure unchanged. Shared global and edge noise components encourage neighboring chunks to choose compatible modes. On AntMaze-Giant, success rises from 65.2% to 76.6%, with nearly unchanged planning time; CDGS scores higher but takes longer.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.