Multi-Junction Traffic Light Control System with Reinforcement Learning in Sunway Smart City

This paper presents a reinforcement learning approach to address urban traffic congestion with traffic light control optimization. The proposed system leverages Q-learning and Markov Decision Processes (MDPs) to develop an adaptive control framework capable of adjusting traffic signal timings dynamically. A neural network model is employed to estimate Q-values, representing cumulative rewards for various control actions. The training process involves interactions between the agent and a simulated traffic environment using the Simulation of Urban Mobility (SUMO) tool. Experimental results showcase the efficacy of the RL-based approach in mitigating congestion and improving traffic flow. This work focuses on the case study of a network of junctions that contribute to the main traffic flow in the township of Bandar Sunway.