Reproduce “VehicleArena: A Realistic Urban Environment for Multi-Agent 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
Jie Yang, Jiajun Chen, Jiazheng Zhou, Mianqiu Huang, Yining Zheng, Yuxin Wang, Xipeng Qiu
VehicleArena evaluates agents driving amid traffic and responding to passenger requests in a shared simulation. It combines cabin service, driving, and reactive vehicles, measuring arrival separately from service and driving scores. In 32 Multi-Agent tasks, Qwen3.8-Max had the highest arrival rate, 65.6%; GPT-5.6-Sol scored 94.0 on Cabin and 87.2 on Drive but arrived on 9.4%. Every tested ego policy lowered surrounding-vehicle arrival versus SUMO in matched runs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.