Platooning Control Of Autonomous Vehicles Based On Deep Reinforcement Learning

Traffic systems are currently facing problems of road congestion and poor vehicle safety.In order to improve road utilisation,vehicle efficiency,save travel time and energy consumption and ensure safe driving,vehicle platoon control technology is required.This technology controls the vehicle spacing and speed so that the vehicle platoon achieves a predetermined desired value.This paper addresses the problems of model uncertainty,poor fuel economy and unpredictable driver behaviour in hybrid driving in the optimal control of self-driving vehicle platoons,and proposes the use of deep reinforcement learning methods to achieve optimal control of vehicle platoons in following.The main research elements of this paper are as follows.Firstly,an intelligent transportation experimental platform is built,which consists of intelligent vehicles,visual information localisation system and wireless communication system,and independent design of hardware and software.The platform can be applied to the study of deep reinforcement learning algorithms to optimally control vehicle platoons,providing experimental guarantees to validate the algorithms.Secondly,an optimised control algorithm based on deep reinforcement learning is proposed to address the model uncertainty problem in self-driving vehicle platoon.A follow-the-leader model for the cooperative adaptive cruise system of self-driving vehicles is established,and a Markov decision process is designed by defining state variables,action variables and reward functions.A deep deterministic policy gradient algorithm is used to train the longitudinal following control of the self-driving vehicle platoon and to find the optimal control policy.The algorithm enables self-driving vehicle platoons to maintain the desired vehicle distance and speed.The simulation results verify the feasibility of the control algorithm.Finally,a safety-enhanced learning algorithm is proposed to solve the optimal control problem of following with safety constraints in mixed vehicle platoons,taking into account the mixed driving of manned and self-driving vehicles in real traffic environments.A model is developed for the following of a mixed vehicle platoon and a Markov decision process with constraints is designed.To address the applicability of safety control in vehicle platoon,the reward function and constraint function are designed,and an optimised control algorithm based on safety reinforcement learning is used to obtain an optimal learning strategy that respects the safety constraints,so as to achieve the control objective of maintaining the desired vehicle distance and speed in the platoon of hybrid driving vehicles,while ensuring the safety of the vehicles.Simulation results verify the effectiveness of the control algorithm.