Reproduce “Designing Versatile Samples for Learned Trajectory Scoring” 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

Yaguang Li, Jiaru Zhang, Chuheng Wei, Can Cui, Ziran Wang

The authors train a driving-policy scorer with scene-specific perturbations of logged human paths toward drivable-area edges and lead vehicles, creating boundary-focused examples. The planner remains frozen, and inference candidates are unchanged. On NAVSIM-v2 navtest, designed samples score 90.41 EPDMS versus 90.12 for plain training and 90.22 for matched random samples; EPDMS combines rule-compliance gates and trajectory-quality terms. This modest gain is benchmark-specific.

Summary by alphaXiv.

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