Exploring Urban Traffic Dynamics: Introducing a Benchmark Map for Comprehensive Testing and Evaluation

Abstract Traffic signal control plays a crucial role in managing traffic flow and alleviating congestion on urban roads. This study proposes a deep reinforcement learning (DRL) approach to optimize traffic signal control and reduce traffic congestion in urban environments. Leveraging Perceiver transformers and a deep neural network, the approach uses traffic flow data — including speed, vehicle arrivals, and other relevant metrics — to enhance signal regulation. The DRL framework is based on the Q-learning algorithm and operates without relying on specific traffic models or rules. To evaluate different traffic control strategies, we introduce a benchmark map of Žilina city as a testing ground. Our method establishes a robust framework for optimising urban traffic, achieving significant improvements in traffic flow efficiency and congestion reduction, as demonstrated through simulations on this map.