On the Applicability of Reinforcement Learning in the Traffic Flow Control Problem

• Published in: Systems of Signals Generating and Processing in the Field of on Board Communications (Online) pp. 1 - 6
• Main Author: Gorodnichev, Mikhail
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.03.2025
• Subjects: Adaptation models; data-driven control; Flow production systems; neural networks; Regulation; Reinforcement learning; Safety; Switches; Traffic control; traffic light control; traffic management; traffic simulation; Training; Transportation; Urban areas
• ISSN: 2768-0118
• DOI: 10.1109/IEEECONF64229.2025.10948029

This article presents the development of a traffic flow regulation method based on an adaptive approach that accounts for changing characteristics of traffic conditions. The study explores key reinforcement learning techniques utilizing Q-Learning to enhance traffic management. Simulations of traffic flow control were conducted in the SUMO simulation environment, focusing on cyclic switching methods. The article outlines the implementation of an environment management strategy and the application of reinforcement learning algorithms via the SUMO API. Two reinforcement learning techniques were selected for traffic flow regulation, and extensive training was conducted with various hyperparameters to identify optimal configurations. A comparative analysis of traditional cyclic switching methods and data-driven approaches is provided, highlighting the potential advantages of adaptive methods for improving traffic efficiency and safety in urban environments. The findings suggest that reinforcement learning can significantly enhance traffic management systems in real-time applications.