A multi-agent reinforcement learning based approach for intelligent traffic signal control

This study addresses the intricate challenges of urban traffic congestion by presenting a novel Multi-Agent Reinforcement Learning (MARL) approach. In response to the critical need for adaptive traffic management solutions in multiple intersections networks, the proposed model integrates a dual-tiered reward system, encompassing personalized and mutual rewards, alongside the incorporation of the replay memory concept. This integration aims to thoroughly evaluate traffic light contributions and enhance traffic signal control strategies. The carefully defined parameters within the reward function are closely aligned with overarching system objectives, specifically targeting the minimization of congestion, delays, and emergency response times. Through simulated scenarios featuring diverse traffic conditions, the proposed MARL model exhibits promising outcomes, demonstrating remarkable adaptability and efficiency even under high traffic volumes. Comparative results with traditional methods, including fixed-time, actuated control, green wave strategy and distributed multi-agent Q-learning method, show the model's robust learning and adaptive capabilities, resulting in substantial reductions in queue lengths and delays. These results validate the model's effectiveness in optimizing traffic flow and efficiently managing congestion, thus making a significant contribution to the advancement of intelligent traffic systems.