Reproduce “Reinforcing End-to-End Autonomous Driving Safety via Negative-Aware Planning in a Unified World-Action Model” 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
Zihao Zhang, Xinyu Zhang, Bing Zeng, Xingyu Chen, Zhenning Li, Bin Yu, Ziyuan Pu
NegWAM trains a world-action model to predict driving actions while learning from offline-mined unsafe trajectories near expert paths. It combines a penalty moving predictions away from these paths with a scene-conditioned objective ranking expert paths above failures; flow-matching imitation remains. On NAVSIM v2 navhard, it scores 40.4 EPDMS versus 38.8 for BeyondDrive, a benchmark result, not a real-world safety guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.