An LLM-Powered Cooperative Framework for Large-Scale Multi-Vehicle Navigation
Yuping Zhou, Siqi Lai, Jindong Han, Hao Liu
TL;DR
CityNav introduces a hierarchical, LLM-powered framework for city-scale multi-vehicle dynamic navigation by coupling a global traffic allocation agent with regional local navigators. It deploys a cooperative reasoning optimization with dual rewards to align local decisions with network-wide objectives, and validates the approach across four real urban road networks using SUMO simulations, outperforming nine baselines in throughput and congestion mitigation. The framework demonstrates strong generalization in zero-shot transfers (e.g., NYC to Chicago) and robustness to increasing demand, highlighting the potential of domain-tuned LLMs for scalable, cooperative urban routing. These findings suggest that structured, multi-level reasoning can unlock efficient, city-wide vehicle coordination without prohibitive training costs.
Abstract
The rise of Internet of Vehicles (IoV) technologies is transforming traffic management from isolated control to a collective, multi-vehicle process. At the heart of this shift is multi-vehicle dynamic navigation, which requires simultaneously routing large fleets under evolving traffic conditions. Existing path search algorithms and reinforcement learning methods struggle to scale to city-wide networks, often failing to capture the nonlinear, stochastic, and coupled dynamics of urban traffic. To address these challenges, we propose CityNav, a hierarchical, LLM-powered framework for large-scale multi-vehicle navigation. CityNav integrates a global traffic allocation agent, which coordinates strategic traffic flow distribution across regions, with local navigation agents that generate locally adaptive routes aligned with global directives. To enable effective cooperation, we introduce a cooperative reasoning optimization mechanism, in which agents are jointly trained with a dual-reward structure: individual rewards promote per-vehicle efficiency, while shared rewards encourage network-wide coordination and congestion reduction. Extensive experiments on four real-world road networks of varying scales (up to 1.6 million roads and 430,000 intersections) and traffic datasets demonstrate that CityNav consistently outperforms nine classical path search and RL-based baselines in city-scale travel efficiency and congestion mitigation. Our results highlight the potential of LLMs to enable scalable, adaptive, and cooperative city-wide traffic navigation, providing a foundation for intelligent, large-scale vehicle routing in complex urban environments. Our project is available at https://github.com/usail-hkust/CityNav.