Enhancing Reinforcement Learning-Based Adaptive Traffic Signal Control in Emerging Mixed Traffic Environments Through CAV Data: A Simulation Study Considering Driver Behaviour

In recent scholarly discourse, it has been noted that connected and automated vehicles (CAVs) have offered substantial developmental potential and a variety of implementation possibilities for the refinement and advancement of adaptive traffic signal control (ATSC). However, as the market penetration rate of CAVs is still relatively low, it is essential to consider the use of existing, cost-effective detectors that can form an integral component of the ATSC system. Additionally, it is crucial to consider drivers' behaviours in the context of emerging mixed traffic environments in order to reflect the realities of traffic flow in simulations due to the distinction between human-driven vehicles and CAVs. To address these issues, an ATSC algorithm was proposed to optimize signal timing to improve safety and operational performance at isolated intersections. The proposed algorithm leveraged real-time Q-learning with loop detector and CAV data to obtain optimized green time, while a driven-behaviour model was introduced to describe human factors in mixed traffic environments. Numerical studies were conducted using simulation of urban mobility (SUMO) to evaluate algorithm performance and investigate the influence of different factors. The results indicate that the proposed algorithm has significant practical value in simultaneously improving safety with a demonstrated reduction in the conflict rate ranging from 28.6% to 72.7% and operational efficiency with a drop in the waiting time ranging from 12.6% to 61.4%, compared to traffic-actuated control. Moreover, it is low-cost and adaptable, and can be continuously updated with real-time driving data while also serving as a layer in next-generation high-definition maps.