Microscopic simulation based study of pedestrian safety applications at signalized urban crossings in a connected-automated vehicle environment and reinforcement learning based optimization of vehicle decisions

This study develops a vehicle-pedestrian safety application for signalized urban crossings in a connected-automated vehicle (CAV) environment and establishes a microscopic simulation environment to implement the safety application using a highly flexible open-source simulation tool. A widely-used surrogate safety measure (SSM) namely, post-encroachment time (PET), is used to capture the number of conflicts. A reinforcement learning algorithm is used to train the CAV agents to determine the optimal timing to cross the intersection with the consideration of pedestrian safety. A variety of real-world datasets are used to calibrate and validate the simulation environment. The simulation results show that the pedestrian safety application can significantly reduce the number of potential conflicts, and the reinforcement-learning-trained CAV agents have demonstrated lower average travel times when crossing intersections.