Reproduce “DriveAnchor: Progressive Anchor-based Flow Learning for Autonomous Driving 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
Limin Yan, Haoyun Tang, Yutao Qiu, Hongqing Liu, Zhandong Mei, Haoyu Xu
DriveAnchor uses a shared library of recorded trajectories to generate route-compliant candidates and adapt a deterministic flow-matching planner offline from static-collision feedback. It shifts starting trajectories toward requested corridors, then fine-tunes corrections using reward queries along nearby trajectory directions. On 10,000 held-out scenes, fine-tuning lowers candidate static-collision rates and minADE, but these measures do not establish collision-free execution or cover dynamic-agent collisions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.