Joint Optimization of Vehicle and Pedestrian Traffic Signals Using Multi-Objective Deep Reinforcement Learning

초록

Traffic signal control (TSC) at urban intersections is crucial for optimizing vehicle traffic flow and ensuring pedestrian safety. Advances in Internet of Things (IoT) and Internet of Vehicles (IoV) technologies have significantly improved traffic monitoring. However, most existing TSC studies primarily focus on optimizing vehicular traffic flow metrics such as waiting time and queue length, often overlooking crucial factors like lane fairness, emergency vehicle priority, and pedestrian convenience and safety at crosswalks. This paper proposes a novel deep reinforcement learning (DRL)-based TSC framework that jointly optimizes vehicular and pedestrian requirements. Two algorithms are introduced to address these multi-objective goals: DFASD (Dynamic Feasible Action Set Derivation), which guarantees pedestrian crosswalk requirements by leveraging a set of feasible actions during action selection, and ACSCS (Adaptive Crosswalk State Combined System), which integrates crosswalk sub-states into the state representation. Simulation results demonstrate that the proposed methods outperform conventional DRL-based dynamic signal control and cycle-based approaches, achieving superior performance in both vehicle traffic flow and pedestrian crosswalk management. These results underscore the potential of the proposed framework to effectively balance vehicular and pedestrian needs, enhancing urban intersection management.

키워드