Optimizing Traffic Signal Control Using Reinforcement Learning for Enhanced Traffic Flow Efficiency

In this paper, the techniques used to optimize traffic light timing in urban environments are discussed. It remains a significant challenge, especially with the increasing complexity of current traffic systems. Tabular Q-learning was the mainstay of early methods from the 1990s, but it was constrained by narrow state spaces and simple traffic measures, which frequently led to less-than-ideal management techniques. [Recent advancements in deep reinforcement learning (DRL) offer new possibilities by combining deep learning with RL to estimate Q-values more effectively. However, most existing studies still focus on simple intersection models, lack comprehensive cycle timing, and do not utilize the latest advancements in DRL, such as dueling deep Q networks. By creating a model that can handle intricate intersection scenarios with several phases and high-dimensional action spaces, this research seeks to overcome current shortcomings. Smooth transitions between traffic signal phases are integrated with advanced value based deep reinforcement learning techniques in the proposed approach. Comprehensive simulations are used to assess how well the model performs in terms of timing optimization for traffic control systems, showing enhanced performance under certain conditions. This research leverages advanced Dueling DQN techniques to address the complexities of multi-phase intersection management, distinguishing it from existing models that primarily focus on single-phase intersections [16].