A Research On Performance Limit Testing Technology For Automatic Driving System

The automatic driving system can significantly improve the traffic safety;promote the transformation of the automobile industry and travel mode.There is a research boom about it all over the world.Since the automatic driving system is directly related to traffic safety,it is very important to detect its performance limit quickly and accurately.The traditional genetic algorithm(GA)based evolution test strategy(GA test strategy)can not compare the advantages and disadvantages of offspring in advance when detecting the performance limit.Therefore,it is blindness when generating the offspring,resulting in low test efficiency.Aiming at the problem,this paper proposes an index to evaluate the effectiveness of offspring without testing,and then uses it design full crossover and multiple mutation operators based on GA,for improving the effectiveness of the offspring.The statistical characteristics of these two operators and the convergence of the improved algorithm is analyzed in theory.In order to effectively implement the test strategy,aiming at the low automation of the traditional test platform,this paper integrates an automatic test platform.Firstly,considering automatic driving systems are high order nonlinear system,it is difficult to evaluate its performance in different scenarios through analytic functions.This paper constructs a tree model to classify the scenario elements,then uses Analytic hierarchy Process calculate the complexity of scenarios,so as to estimate their effectiveness.Furthermore,in order to improve the probability of generating high-quality offspring,on the basis of single point crossover and simple mutation operators,the full crossover and multiple mutation operator that can optimize the offspring according to the complexity is designed with introducing the concept of brother competition,and the full process of Improved GA(IGA)test strategy is given.By comparing the mathematical expectation of the complexity of offspring generated by the above two crossover/mutation operators,this paper also proves theoretically that full crossover and multiple mutation operators are helpful to improve the complexity of offspring.To analyze the effect of IGA test strategy,based on GA convergence theory,the global convergence of IGA is proved by analyzing state probability transition matrix of its key operators.Since it is difficult for a certain software to realize scenario construction,data processing and some other functions at the same time,this paper integrates simulation test platform with selecting Pre Scan as scenario simulation software,MATLAB as data processing software.Then,based on the characteristics of IGA test strategy and interface of scenario editing provided by Pre Scan,the overall process of automatic test is designed,and automatic 3D scenario reconstruction technology is developed,realizing the automation of test.Besides,based on COM technology,the function of automatically generating test reports is realized.At last,the performance limit of the autonomous parallel parking system to avoid collision with the road edge is tested by using the test strategy and platform proposed in this paper.The results show that complexities of scenarios calculated by the method proposed in this paper can effectively predict scenarios’ qualities.Under the guidance of brother competition,full crossover and multiple mutation operators can prominently speed up the process that offspring evolve to the high complexity.It makes IGA test strategy significantly improve test efficiency.