Abstract

This paper presents the design, development, and experimental evaluation of a Reinforcement Learning (RL) based Adaptive
Traffic Signal Control (ATSC) system for intelligent urban traffic management. The proposed approach formulates traffic signal
optimisation as a Markov Decision Process (MDP) and applies Deep Reinforcement Learning (DRL) algorithms specifically
Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C) to enable real-time adaptive
decision-making under dynamic traffic conditions. The system is organized into four integrated modules covering user
interaction and visualization, RL algorithm execution, reward-based feedback, and traffic simulation. Experimental outcomes
show that RL agents, particularly PPO, consistently surpass conventional fixed-time and actuated control methods across key
metrics including average waiting time, queue length, throughput, and estimated emissions. Network-wide efficiency is further
improved through multi-agent coordination using a Centralised Training and Decentralised Execution (CTDE) strategy. The
framework provides a scalable, modular, and extensible platform for future research and smart city integration.