Multi-agent deep reinforcement learning-based decentralized adaptive traffic signal control for corridor-level safety optimization

This paper presents a novel contribution to traffic signal control by developing a multi-agent decentralized deep reinforcement learning algorithm based on proximal policy optimization for corridor-level adaptive traffic signal management with a focus on real-time safety optimization. Unlike most existing methods that prioritize mobility alone or operate at isolated intersections, our approach integrates a real-time safety measure, the return level, derived from extreme value theory crash risk models, into the learning framework to optimize safety and mobility along signal corridors. Simulation results demonstrate significant improvements in both safety and efficiency, making this a promising direction for future innovative traffic systems.