Design And Implementation Of Unmanned Vehicle Control System Based On Deep Reinforcement Learning

Along with the rapid development of China's economy and the substantial improvement of people's living standards,the number of small family cars in China has grown rapidly,and the road traffic problems caused by this have become increasingly serious.With the growth of market demand and the maturity of technology,the research on unmanned vehicles has developed rapidly in recent years.In this paper,we take unmanned vehicles as the research object,and carry out research on the automatic driving task of unmanned vehicles based on deep reinforcement learning.This paper aims to realize an experimental model of an unmanned vehicle with automatic driving function,and use this model to explore and verify the relevant theories and techniques of automatic driving of unmanned vehicles.In the implementation of the unmanned vehicle control system,this paper uses the deep reinforcement learning theory.When using the deep reinforcement learning theory to realize the control system of the unmanned vehicle,a lot of training is needed.This is very difficult to implement in actual operation.Therefore,this paper proposes the training of the unmanned vehicle control system using the simulation platform.Because there is a difference between the simulation platform and the actual environment,the neural network model after training cannot be directly applied to the unmanned vehicle.Aiming at this problem,this paper proposes a technique of using image semantic segmentation to eliminate the difference between the simulation platform and the real environment,so that the model trained on the simulation platform in this paper can be directly applied to the unmanned vehicle task.The main research results of this paper can be summarized as follows:First,this paper designs and implements an image semantic segmentation system.In the design and implementation process of image semantic segmentation system,this paper fully considers the application environment after the system.In practical applications,the on-board computer performance used in this article cannot be compared to the mainframe computer in the lab.Therefore,in the design and implementation of the control system,this paper requires that the system's requirements for computing power should be kept at a low level,and the power consumption required during operation should not be too high.At the same time,considering the high response speed of each system in the unmanned vehicle task,the image semantic segmentation system of this paper has good real-time performance.Second,this paper designs and implements the control system of the unmanned vehicle.The system can make decisions based on the input image,control various actions of the unmanned vehicle,and ensure the safe and smooth operation of the unmanned vehicle.In the design of the system,this paper is mainly based on deep reinforcement learning.In this paper,a large number of training processes required by the network are put on the simulation platform,which greatly shortens the training time and difficulty.In this paper,the hardware performance of the computer is fully utilized in the training,and the parallel asynchronous training method is adopted,which reduces The space requirements required for training increase the speed of training.Third,this paper designs and implements the hardware structure of the unmanned vehicle,including the design of the mechanical structure and the design of the control circuit structure.In the hardware design of the unmanned vehicle,this paper strives to improve the reliability and safety of the hardware facilities,and uses an independent suspension system to ensure the smooth and safe operation of the unmanned vehicle on the hardware.And this paper uses the mature PID algorithm as the underlying control algorithm to ensure the precise control of the unmanned vehicle control system.Fourth,this paper designs and conducts a large number of experiments to show the excellent performance of the algorithm and verify the feasibility of the proposed method.Finally,the design implementation work and research content of this paper are summarized,and the shortcomings in the design and implementation are pointed out,which explains the next research direction of unmanned vehicles.