A Survey on RL-Based Approaches for Traffic Signal and Joint Vehicle-Signal Control

With the rapid increase in urban vehicular traffic, the need for intelligent traffic management systems has become critical to improve road safety, alleviate congestion, and reduce carbon emissions. Traffic Signal Control (TSC) systems have advanced significantly, leveraging technologies to optimize signal timing and coordination. Data-driven approaches are transforming transportation, robotics, IoT, and power systems. Integrating these techniques into traffic management is essential to tackle modern urban mobility challenges. This paper surveys recent applications of deep reinforcement learning (RL) in traffic control, focusing on TSC and vehicle speed control (VSC) to optimize urban traffic flow and safety. We provide an overview of core deep RL concepts, examine RL models for TSC and vehicle speed control, and highlight their advantages. The survey also explores key RL-based TSC components-state representation, action spaces, and reward structures-that affect performance. Additionally, we discuss challenges and innovations, emphasizing advancements and future pathways for reinforcement learning in vehicle-light coordination systems.