Deep Reinforcement Learning Approach for Traffic Light Control and Transit Priority

This study investigates the use of deep reinforcement learning techniques to improve traffic signal control systems through the integration of deep learning and reinforcement learning approaches. The purpose of a deep reinforcement learning architecture is to provide adaptive control via a reinforcement learning interface and deep learning for the representation of traffic queues with regards to signal timings. This has driven recent research, which has reported success in the use of such dynamic approaches. To further explore this success, we apply a deep reinforcement learning algorithm over a grid of 21 interconnected traffic signalized intersections and monitor its effectiveness. Unlike previous research, which often examined isolated or idealized scenarios, our model is applied to the real-world traffic network of Via “Prenestina” in eastern Rome. We utilize the Simulation of Urban MObility (SUMO) platform to simulate and test the model. This study has two main objectives: ensure the algorithm’s correct implementation in a real traffic network and assess its impact on public transportation, incorporating an additional priority reward for public transport. The simulation results confirm the model’s effectiveness in optimizing traffic signals and reducing delays for public transport.