Enhancing Traffic Efficiency through Real-Time Traffic Signal Control Using Deep Q-Learning

This research manuscript delves into the employment of advanced technologies for the enhancement of traffic efficiency, leveraging real-time data and deep reinforcement learning methodologies. We focus on the formulation of a specialized Deep Q-learning agent, capable of effectively managing traffic light operations at an isolated junction, utilizing real-time GPS data. The amalgamation of deep neural networks with the Q-learning component of reinforcement learning enables the agent to refine traffic signal control strategies. The Deep Q-Network (DQN) algorithm is utilized, in combination with replay memory and a target network, to bolster and stabilize the learning journey. The agent’s training process employs Q-learning with experience replay within the traffic simulator SUMO, fostering the production of traffic signal control policies. The suggested method aspires to curtail traffic congestion and reduce waiting periods by facilitating intersections to make judicious decisions premised on their real-time status. The metrics are queue length and reward of the agent which are carefully monitored for the best result.