Reproduce “Learning Agent-Centric Vision-Language-Action World Models for Autonomous 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

Guoqing Wang, Pin Tang, Xiangxuan Ren, Chao Ma

AgentWorld represents nearby traffic participants as grounded agent tokens, combining geometric and visual features into addressable states for trajectory and action planning. On nuScenes, it achieves 0.27 m average trajectory L2 error and 0.08% collision rate, versus 0.49 m and 0.23% for global latents. Bench2Drive simulation results do not establish real-road performance; grounding requires geometric supervision.

Summary by alphaXiv.

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