Study On The Test And Comprehensive Evaluation Of Conditional Autonomous Vehicles

The test and evaluation of automated vehicles is crucial for the development of autonomous driving.Before the real driving road test of automated vehicles,it is necessary to carry out complete tests on the automatic driving system,including simulation test,discrete scene test and continuous dynamic traffic scene test of the close site.It is guaranteed that most system defects are found before the road test,Ensuring the safety and testing efficiency of open road testing.Therefore,this paper studies on the test and comprehensive evaluation methods of L3 automated vehicles,including the following contents:(1)For the L3-class autonomous driving systems,Monte Carlo simulation based on natural driving data was studied.Using the kernel density estimation to obtain the probability distribution of the scene parameters.MH sampling and IS sampling are used to generate common test cases and key test cases respectively,which are used for Monte Carlo simulation test to estimate the performance of the autopilot system in natural driving scenarios.Based on the simulation result,key test cases are extracted using Euclidean distance clustering for further simulation testing and site testing.Finally,the method is applied to the simulation test of the front brake condition of the ACC function.(2)for the pre-testing of L3-class autonomous driving systems at closed-site,a combined test scenario generation method based on vehicle position and motion direction of a high-level vehicle combination group is proposed.The method is based on the most complicated vehicle combination group in designated road traffic environment,using the PICT tool to combine the motion directions of the jamming vehicles that affect the main vehicle motion to generates test scenarios with different potential dangers.And based on the scene screening principle,Filter out valuable test scenarios.Finally,the method is applied to the test scenario design of road sections and intersections,and the test cases are designed by parametric design of the scene.Co-simulation tests were performed on the generated test cases based on PreScan,CarSim,and Matlab/Simulink to verify the usability of the test cases.(3)For the L3 automated driving system,which has the characteristics of continuous operation for a long time,a continuous dynamic traffic scene generation method for closed sites is designed.By analyzing the traffic elements of the autopilot function behavior,the method combines these elements based on the hierarchical task form to generate continuous scenarios.According to the dynamic and interactive scene design method and the traffic complexity calculation model,the continuous scene is optimized and the elements are adjusted.To form continuous and dynamic traffic scenarios that can be used for testing.Finally,Development of GUI interactive simulation test platform based on Matlab,Testing the continuous dynamic traffic scenarios on the platform for urban autonomous driving systems.(4)For the evaluation of L3 automatic driving,a general evaluation system and comprehensive evaluation model were established.For the specific automatic driving function,the evaluation index selection and screening process was designed.Comprehensive weighting of indicators using AHP entropy method,and a grey comprehensive evaluation model and a BP neural network comprehensive evaluation model were established for the evaluation of automatic driving function.The evaluation method is applied to the comprehensive evaluation of the high-speed automatic driving system.