Traffic Signal and Autonomous Vehicle Control Model: An Integrated Control Model for Connected Autonomous Vehicles at Traffic-Conflicting Intersections Based on Deep Reinforcement Learning

Abstract

This article studies effective and safe control problems for connected autonomous vehicles at crowded and traffic-conflicting intersection. A novel control model integrated traffic signal control and autonomous vehicle control called signals and vehicles integrated control (SVC) is proposed to improve the efficiency and safety of vehicles through intersections, where the speed of each vehicle is controlled under the constraint of the controlled traffic light. The traffic signal timing model which determines the duration of each signal phase is composed with a flow predictor and a timing model. The autonomous vehicle control model is designed based on the idea of human-machine cooperation and Soft-Actor-Critic algorithm to ensure autonomous vehicles on the signal-allowed directions travel through the intersection efficiently and safely. The experiment results show that compared to commonly used traffic signal control models, pure autonomous vehicle control models and traditional traffic signal control combined with autonomous vehicle control models, the proposed SVC achieves better asymptotic performance in a crowded intersection under the comprehensive metric which considering both efficiency and safety.