On Max Pressure Urban Traffic Control with Learning

Managing the traffic system is a multifaceted challenge, imposing various solutions to prevent congestion. Traffic management aims to create an efficient control system with numerous advantages, including reduced fuel emissions, lower stress levels, and enhanced driving discipline. Max Pressure stands as one of the techniques for traffic control, designed to adapt and enhance network throughput by considering pressure, defined as the difference between upstream and downstream queue lengths. Recent developments have yielded improved performance in Max Pressure through learning. In this adaptive control mechanism, the concept of Max Pressure is integrated as a reward function within reinforcement learning. This study introduces the “Max Pressure learning” concept, revealing its potential for creating adaptable traffic controllers. By modifying the pressure equation and transforming it into a performance metric, the original Max Pressure concept effectively responds to diverse traffic conditions. This paper showcases research findings demonstrating the surprising effectiveness of pressure as a control mechanism in the context of the learning process. Ultimately, using pressure enables controllers to efficiently alleviate gridlock situations in traffic, highlighting its promising role in optimizing traffic management and enhancing overall traffic system performance.