Reproduce “Safety4DM: A Vision-Language Framework for Risk-Aware Dynamic Maps in Vehicle-Road-Cloud Collaborative 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
Zhengshu Zhou, Shaoyu Wang, Lu Tao, Xinhui Hu, Jiaqi Gu, Tingting Zhao
Safety4DM converts road images into structured hazard records and enables dynamic-map queries through SafeStep, whose deterministic toolbox retrieves entities and computes values before an LLM forms an answer. On simulated-scene questions, full SafeStep achieves 96.3% exact-result accuracy, versus 34.8% for one-prompt prompting and 33.7% without the toolbox. Fine-tuned Qwen3-VL-4B reaches 89.83% joint risk-perception accuracy; neither result establishes safer live driving.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.