AI-Powered Traffic Signal Control: A Deep Reinforcement Learning Framework for Urban Congestion Reduction

Abstract

Urban traffic congestion remains a serious issue as it enhances fuel consumption, carbon emissions, and travel time. Traditional traffic signal control systems often cannot adapt to varying traffic conditions because of their fixed or rule-based timings, which results in inefficiency. To achieve maximum urban mobility, this research explores AI-based traffic signal regulation using machine learning and reinforcement learning techniques. An AI agent continuously learns and adapts signal timings from sensor and camera readings of real-time traffic in this work, defining traffic signal optimization as a sequential decision-making task. A deep reinforcement learning framework is applied to minimize waiting times for vehicles and maximize the efficiency of traffic flow, and it utilizes approaches such as Proximal Policy Optimization (PPO) and Deep Q-Networks (DQN). On real-world traffic data, simulations are performed using SUMO (Simulation of Urban Mobility), comparing AI- based methods with traditional fixed-timing and actuated traffic control systems. The findings indicate that AI-based solutions enhance vehicle throughput, reduce average delay times, and significantly alleviate congestion. Adaptive traffic signals also enhance pedestrian safety and emergency vehicle flow. Based on the findings of the study, AI-powered traffic light control offers a smart and scalable solution for urban mobility issues, and it can even be integrated into smart city infrastructures. Decentralized traffic management through multi-agent reinforcement learning and issues related to practical deployment will be explored in future studies.