Deep Deterministic Policy Gradient Algorithm based Lateral and Longitudinal Control for Autonomous Driving

The traditional static path planning algorithm for automatic driving has the problems of insufficient robustness and unstable body control under complex road conditions. It can make full use of the reinforcement learning related technology to carry out vehicle decision-making and control, plan the dynamic obstacle avoidance route and enhance the driving safety. In this paper, a vehicle control algorithm is designed on the TORCS (The Open Racing Car Simulator) simulation platform combined with DDPG (Deep Deterministic Policy Gradient) algorithm. Combined with actor critic algorithm, experience playback and independent target network are added to improve the effect of deep reinforcement learning, with better robustness. And by designing a reasonable reward function, the car can be more stable in the driving process. In the simulation experiment, the algorithm is verified. The experimental results show that the algorithm designed in this paper can make the simulation car get more stable and effective learning in less training time, that is, the research idea of this paper is feasible.