Multi-Agent Deep Reinforcement Learning for Efficient Vehicular Mobility via Federated Fog Computing

Optimizing vehicle waiting time is essential for an intelligent intersection management system as it directly influences traffic flow, public safety, and overall network performance. Conventional emergency vehicle management methods rely on static rules that are unable to adapt to dynamic traffic conditions. This paper presents MADQL-FFC, a multi-agent deep Q-learning framework combined with federated fog computing for real-time traffic signal control at the intersection that prioritizes emergency vehicles. In this framework, deep Q-network agents are deployed at individual intersections to make localized decisions, while a federated fog computing architecture is used in information sharing among neighboring agents for coordinated decision making. This helps in decentralized, low-latency coordination across the traffic network without the need for centralized control. The proposed approach is evaluated using realistic traffic simulations based on OpenStreetMap data of Gwalior City, India, implemented in the Simulation of Urban Mobility (SUMO) simulator. Comparative results against state-of-the-art methods demonstrate that MADQL-FFC achieves reductions in waiting time for all vehicle categories. Overall, the proposed model achieves a performance improvement of 5% to 32%, 12% to 44%, 4% to 46%, and 13% to 38% in average waiting time, average vehicle speed, average queue length, and average throughput respectively, when compared to state of the art methods.