Decision-Making Models for Autonomous Vehicles at Unsignalized Intersections Based on Deep Reinforcement Learning

Abstract

Decision making at unsignalized intersections is a critical challenge for autonomous vehicles. Navigating through urban intersections requires determining the intentions of other traffic participants. Solving this complex decision-making problem with traditional methods is difficult. To eliminate conflicts at intersections, this paper introduces several deep reinforcement learning algorithms. This research modeled the behavior of drivers at these intersections. Using this, reward functions were designed, and a meta exploration deep deterministic policy gradient was reorganized. Finally, a novel time twin delayed deep deterministic policy gradient algorithm was developed that considered prediction factors. The Carla-Gym simulation platform was used to build an unsignalized intersection model. The experimental results show that the improved deep reinforcement learning method performed better for navigating autonomous vehicles through unsignalized urban intersections.