Reproduce “End-to-end Interpretable Neural Motion Planner” 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

Wenyuan Zeng, Wenjie Luo, Simon Suo, Abbas Sadat, Bin Yang, Sergio Casas, Raquel Urtasun

NMP learns a time-indexed cost over ego-vehicle positions from LiDAR and a high-definition map, then selects the lowest-cost physically feasible trajectory from sampled candidates. It also predicts other vehicles’ boxes and future motion. On logged driving data, its three-second potential collision rate was 0.78%, versus 2.21% for a manual-cost planner using the same sampler. This is not a safety guarantee.

Summary by alphaXiv.

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