Synergistic Reinforcement Learning Models for Pedestrian-Friendly Traffic Signal Control

Traffic signal control is essential for managing urban traffic, reducing congestion, and minimizing environmental impact by optimizing both vehicular and pedestrian flow. This paper investigates the application of Reinforcement Learning (RL) in traffic signal control within mixed traffic environments, emphasizing the development of a synergistic RL approach, named Advantage Actor-Critic with Maximum Pressure (A2CMP). A2CMP leverages actor-critic techniques in combination with real-time pressure metrics to dynamically adjust traffic signals based on prevailing traffic conditions. Additionally, the paper introduces a pedestrian-friendly phase-skipping mechanism for further enhancing the efficiency of the proposed algorithm in real-world traffic management. Simulation results across diverse traffic scenarios show significant reductions in CO2 emissions and waiting time. Particularly, A2CMP can reduce waiting time by 12% compared to other RL-based algorithms.