Intersection Self-Organization Control for Connected Autonomous Vehicles Based on Traffic Strategy Learning Algorithm

With the rapid advancement of intelligent vehicles and vehicular communication systems, connected autonomous vehicles (CAVs) will run on the road in the foreseeable future. To increase the traffic efficiency of CAVs at intersections, it is necessary to apply a new method to replace the traditional signal time assignment. This paper proposes a general solution for CAVs passing through non-signalized intersections effectively. A novel idea is developed to use a traffic strategy learning algorithm for real-time decision-making. Through an image representation method based on lanes reordering for intersection state description, the convolutional neural network model is adopted. The proposed methods can take full advantage of spatiotemporal resources at the intersection and ensure the rapidity and efficiency for practical applications. Several numerical experiments in different traffic situations are designed to demonstrate the validity of the proposed method.