Reproduce “ProDrive: Proactive Planning for Autonomous Driving via Ego-Environment Co-Evolution” 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

Chuyao Fu, Shengzhe Gan, Zhuoli Ouyang, Yuhan Rui, Xiaowei Chi, Sirui Han, Jiankun Wang, Hong Zhang

ProDrive introduces a framework for proactive autonomous driving planning that tightly couples a query-centric trajectory planner with a BEV world model, enabling ego-environment co-evolution. The system achieves a Predictive Driver Model Score (PDMS) of 86.6 on the NAVSIM benchmark, demonstrating improved safety and efficiency by anticipating future scene evolution and interactions.

Key insights

  • Bidirectional interaction between a planner and a world model, where each informs the other, is crucial for enabling proactive and safer autonomous driving.
  • Injecting rich, planner-aware semantic context (ego tokens) into the world model significantly enhances the accuracy of future scene prediction and improves the collaborative optimization of the system.
  • Maintaining an end-to-end gradient path from future-aware reward signals (e.g., collision avoidance, progress) back to the planner is vital for learning anticipatory behaviors and making globally optimal decisions.

Summary by alphaXiv.

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