Towards smart mobility: coordinated intersection control using multiagent reinforcement learning and visible light communication

Urban traffic management faces growing challenges due to increasing vehicle and pedestrian flows, leading to congestion, delays, and safety concerns. This study proposes an intelligent, decentralized traffic management system tailored for cities like Lisbon, where radial and circular road topologies demand adaptive coordination. The framework integrates Multi-Agent Reinforcement Learning (MARL) with Visible Light Communication (VLC) to optimize signal control at intersections—the main bottlenecks of urban mobility. Deep Reinforcement Learning (DRL) agents exploit VLC data to learn adaptive phasing strategies that respond dynamically to changing traffic conditions and directional imbalances. The urban model consists of modular unit cells, each with five interconnected intersections where circular and radial arteries converge at a central node identified as critical for network stability. VLC enables high-resolution, real-time data exchange between vehicles, pedestrians, and infrastructure, supporting a dynamic request/response mechanism for adaptive phase adjustments and congestion mitigation. Each intersection operates under a decentralized DRL agent that cooperates with neighbors to develop context-aware strategies, enhancing global coordination and responsiveness. Simulation results show that neighbor-aware decision-making significantly improves traffic flow, reducing queue lengths and increasing average speeds, while maintaining fair pedestrian phases. The system effectively prioritizes dominant flow directions under high-demand conditions, ensuring fluidity within multi-intersection clusters and preventing local bottlenecks. By combining MARL with VLC and Connected Vehicle (CV) technologies, this approach offers a scalable and adaptive solution for next-generation smart cities, addressing the complexity of multimodal traffic in dynamic urban environments such as Lisbon.