Grid planner improves autonomous driving safety and comfort with graph rewards
GRID-Planner: Graph-Structured Relational Reward Shaping for Diffusion Planning
RoboticsComputer Vision and Pattern Recognition
Summary
Planning safe and smooth routes for self-driving cars is hard because cars need to react to many other vehicles. The authors created a new method called GRID-Planner that uses a special way of rewarding possible routes based on relationships between vehicles. This method helps the planning system pick safer and more comfortable paths, reducing crashes and sudden moves. Their tests showed substantial improvements in avoiding collisions, staying in lanes, and smoothness of driving.
What this means in practice
- •For autonomous vehicle engineers: Enhance planning algorithms to reduce collisions and improve comfort by integrating relational reward structures into trajectory selection.$Commercial implications: Enables development of safer self-driving car navigation systems with better multi-agent awareness for commercial autonomous vehicle products.
- •For robotics software developers: Incorporate graph-based relational rewards to improve multi-agent planning in autonomous mobile robots navigating dynamic environments.
Authors
Zhao Yang, Zezhong Qian, Yinan Shi, Longjun Liu, Yawei Jueluo
Abstract
Autonomous driving planning requires selecting trajectories that remain safe, rule-compliant, and comfortable under complex multi-agent interactions, yet robust closed-loop decision making in planning-critical scenes remains challenging. Existing diffusion- and RL-based planners often rely on generic rewards and use relational structure mainly for representation, limiting how planning-critical rollouts shape policy improvement. We present GRID-Planner, a graph--reward integrated diffusion planning framework that aligns generative planning with structured relational signals. Specifically, GRID-Planner combines graph-aware importance weighting, a time-indexed relational reward surrogate, and an advantage-weighted diffusion objective with on-policy self-distillation, forming a rollout-return-advantage-update loop without explicit likelihoods. Experiments on nuPlan and interPlan show that, compared with the diffusion-only baseline, GRID-Planner reduces collision rate, lane deviation, and jerk by 35.8\%, 60.8\%, and 27.4\%, respectively, and improves the planning score by 3.4\%. Project website: https://github.com/yangzhaojason/GRID-Planner.