Towards fair lights: A multi-agent masked deep reinforcement learning for efficient corridor-level traffic signal control

Abstract

This study presents an adaptive traffic signal control (ATSC) method for managing multiple intersections at the corridor level by proposing a novel multi-agent masked deep reinforcement learning (DRL) framework. The method extends the hybrid soft-actor-critic architecture to optimize green light timings for intersections across a corridor network, fostering a balance between vehicle flow and pedestrian movements with an emphasis on humanism, fairness, and equality. By integrating an innovative phase mask mechanism, our model dynamically adapts to the fluctuating demand of different transportation modalities by discovering new states or actions that could avoid local optima and achieve higher rewards. We comprehensively test our method using five naturalistic traffic scenarios in Melbourne, Australia. The results demonstrate a significant improvement in reducing the number of impacted travellers compared to existing DRL and other baseline methods. Furthermore, the inclusion of the phase mask mechanism enhances our model's performance through ablation analyses. The proposed framework not only supports a fairer traffic signal system but also provides a scalable, adaptable solution for diverse urban traffic conditions. .