Reproduce “SimVTA: Simulation-Grounded Video-Trajectory Alignment for Driving World Action Models” 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
Changhui Jing, Haochen Tian, Hongchen Li, Tianyu Li, Caojun Wang, Zihao Zhang, Yanhao Liu, Chengen Xie and 3 more
SimVTA trains driving world-action models to plan trajectories and predict videos using simulator-rendered futures matched to candidate paths. It selects valid, high-reward candidates and supervises high-error futures, while reward-based policy learning and a behavior-cloning regularizer train the planner. On NAVSIM v1, scores rise from 86.2 to 91.3 PDMS; this full-framework result is on Epona, not an isolated video-supervision effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.