| Due to many problems such as low efficiency,high cost,short time,and high risk in actual road testing,it is especially important to develop virtual simulation tests for autonomous driving based on computer simulation technology.Virtual simulation tests can effectively meet the testing needs of a large number of autonomous vehicles.The research content of this article is based on BYD Tang SUV hybrid vehicles.Aiming at the actual vehicle test requirements,I built a highly realistic virtual test environment,and developed an automatic multi-dimensional test case generation algorithm based on probability statistical distribution,which can effectively improve test efficiency.The main research contents are as follows:1)fast modeling of the vehicle using an object-oriented method,simplified analysis of the objects of each vehicle subsystem,constructed a three-dimensional visual model of the vehicle and a transverse or longitudinal dynamic model of the vehicle.Through the analysis of experimental data,the error of the engine model obtained by the least squares fitting method is within 5%,and coefficients of the steering system and the brake system model obtained by linear fitting are 18.26 and5.3.Respectively,through the verification of linear braking conditions and steering step conditions,it can be seen that the simulation model can well reflect the dynamic characteristics of actual vehicle.2)In order to meet the large number of test needs of real vehicles,this paper proposes an automatic generation algorithm of virtual test cases based on probability statistical distribution.Firstly,the traffic scene components are analyzed,and a multi-dimensional virtual scene element library is constructed.The scene dynamic fragments are constructed by randomly combining and arranging different elements,and the ACC scene fragment is parametrically described.By analyzing 2150 traffic accidents in a highway traffic accident data set,the probability distribution of accident time,accident weather and accident type is extracted,and the probability distribution of scenario cases is approximated by analogy to provide a basis for test case allocation.The uniformly distributed resampling method is used to automatically generate scene cases randomly,and the probability distribution is used to randomly generate infinite dynamic cases,which can more effectively test the effectiveness and robustness of the algorithm.3)In order to meet the fidelity of the test environment,this paper studied the simulation method of environmental characteristics using high-definition rendering pipeline technology,through the particle simulation system to simulate the rain,snow and fog,and the modeling of the environmental weather,which can simulate many different changing weather environment.The use of Unity3D’s high-definition rendering pipeline technology can more realistically simulate the real environment,including visibility,environment tone,day and night alternation,flood and depth of field,and dynamic blur,so that the constructed virtual environment is closer to the real environment.4)Finally,based on the above work content,a data collection platform and algorithm verification platform were built,Logitech G27 driving simulator was connected with the virtual environment to collect driving data,depth maps,semantic segmentation maps,2D inspection maps and the data of the 3D inspection chart is saved locally,as the algorithm training verification data set.The dynamic model in the virtual simulation environment is simulated,and the algorithm simulation operation of cruise,follow-up and cut-in conditions in the ACC simulation experiment is performed to verify the reliability and efficiency of the platform. |
PDF Full Text Request