Design And Implementation Of Traffic Simulation Back-end Service System Supporting Automated Vehicle Testing

Autonomous driving is a current research hotspot,and many researchers have begun researching autonomous driving algorithms.A core issue in autonomous driving is the reliability of autonomous vehicles in various environments,which is closely related to human and property safety.Therefore,it is undoubtedly very important to do a lot of testing of autonomous driving software.Because of the high economic and time cost of on-road testing,simulation testing is generally used in the early stages of development,which can effectively improve the efficiency of algorithm software development.Most of the traffic simulation software on the market today is a stand-alone version.It is difficult to flexibly accept different driving algorithms,and will reduce the computing efficiency of driving software running on the same machine.There is also some problem such as low cross-platform and cross-language support,insufficient large-scale simulation computing ability and high cost of replacing simulation software.Hence,as a test environment,it can't meet the needs of researchers well.In order to solve this situation,this paper proposes a simulation test platform based on client-server architecture which encapsulates traffic simulation as a network service.It sets up a simulation environment on the server and accepts client vehicle simulation in the form of network interfaces.Considering the needs of autonomous driving algorithms,a set of server and client network protocols are designed.The service system in this paper has the following functions.1.The driving clients can add,delete,and modify vehicles in the simulation environment through the network protocols,which allows different autonomous driving algorithms to access the server at the same time,regardless of the development platform and language.2.The server allows the traffic signal control logic from the traffic light client to access.3.There are user interface that can control time and weather and stable number of server SUMO vehicles.4.There is an interface that client can control the server simulation process to meet the training needs of the reinforcement learning client.5.The visual client can connect to the server and present the traffic video,which is convenient for the observation and monitoring of the test process.The service system of this paper has the following advantages.1.By connecting the autonomous driving algorithm to the simulation test platform,researchers can focus on algorithm research,without the need to manage the setup of the simulation environment.At the same time,the cross-language feature allows researchers to use any language for research code development without the constraints of simulation software.2.Overall,the client-server architecture distributes computing power across different machines,so that large-scale simulations will not affect the computing efficiency of machine learning algorithms.3.Additionally,the simulation test platform is flexible,and can easily accept multiple different autonomous vehicles in the same simulation environment to interact at the same time,without the need to modify the code each time when a new driving algorithm joins.4.Moreover,it has computational scalability from the perspective of the overall system.Regardless of any upgrade of the server,such as replacing simulation software,changing programming languages,using distributed simulation,increasing simulation capacity,increasing map size,etc.,the server interface is fixed and the driving client has no need to make any changes,which greatly reduces the time of code modification and the possibility of bugs.