Reproduce “TrajWAM: Trajectory-Conditioned Future Learning for Vision-Language Driving” 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

Xuerui Fang, Gulimila Kezierbieke, Chen Wangxing, Hui Li

TrajWAM trains a driving planner by reconstructing logged future features and encouraging different predictions for perturbed trajectory queries. A frozen DINOv3 teacher supplies targets, and the auxiliary branch is bypassed at inference. On 220 Bench2Drive routes, TrajWAM scores 91.53 Driving Score and 75.45% strict Success Rate, improving over a matched planning-only control; evaluation uses one seed and simulator sweep.

Summary by alphaXiv.

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