Reinforcement Learning-Based Cooperative Traffic Control System

Urban traffic congestion is an increasingly pressing issue and advanced solutions like intelligent traffic control systems are becoming unavoidable. This paper explores the application of reinforcement learning to enhance traffic flow and reduce congestion. Our goal was to develop a reinforcement learning-based model that adapts to varying traffic conditions in real-time. Several methods are available for the real-time optimization of traffic, ranging from analytical methods to more flexible metaheuristic methods and reinforcement learning-based solutions. Each method lacks either adaptivity or scalability, or the completeness of the global optimization, or the performance requirement is too high. Based on the different requirements it is clearly a challenging task. The focus of our research is on the scalability, and computational efficiency of the model by using a method of sharing information that is similar to a cellular network. Our solution is not just easy to scale but also able to search for the global optimum with a low computational cost. A general model was trained to achieve out-of-the-box usage capability. The control network also can be finetuned for better performance. Our comprehensive analysis showed that smart traffic lights significantly enhance the efficiency of traffic systems, boasting improvements ranging from 10% to 80% compared to traditional and other reinforcement learning-based solutions. These intelligent controllers not only reduce waiting times but also contribute to environmental protection by reducing the carbon footprint.