Reproduce “GRID-Planner: Graph-Structured Relational Reward Shaping for Diffusion 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

Zhao Yang, Zezhong Qian, Yinan Shi, Longjun Liu, Yawei Jueluo

GRID-Planner uses relational scene information to prioritize training rollouts, score candidate trajectories for safety, rule consistency, and comfort, and guide diffusion-policy updates. In nuPlan’s Test14-hard reactive simulation, it scores 71.15 CLS, versus 66.88 for the authors’ locally reproduced Flow Planner, a 4.27-point gain. This comparison is not against Flow Planner’s published 70.42 score, and does not establish real-world performance.

Summary by alphaXiv.

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