Study On Simulation Scenario Generation And Test Method For Intelligent Driving System

Industrial development of intelligent vehicle has become a critical strategic direction and trend under the tendency that electric,intelligent and connected automobile developed quickly.For the reason that intelligent vehicle is composed of several fields such as traditional automotive electronics,information and communication,artificial intelligence,internet and software,integrates end、 pipe and cloud multi-system,and couples human-vehicle-traffic multiple dimensions,brings new issues such as safety,environmental adaptability,human-computer compatibility,it is hard to satisfy the needs to develop intelligent vehicle by using traditional testing method.Therefore,it is necessary to establish an integrated framework aims to evaluate testing and assessing method for intelligent driving system.Consequently,this article designed a platform to collect,generate and accelerate testing scenario based on typical natural driving data in China,to satisfy the needs that intelligent driving system is comprehensively typically and effectively implemented in the testing scenario.Besides,this algorithm has been tested on the simulated platform of test scenarios,which increase the efficiency of simulated tests for intelligent driving.The main contents of this research article are expressed as follows:(1)The constituent factors for existing test scenarios are based on expert experience or test experience,their value is determined by replicating foreign standard working conditions,functional definition or operating area of intelligent vehicles and database of traffic accident caused by drivers,which reflects the defect of the existed test scenarios of Chinese typical driving environment.To solve this problem,this research article implements platform designed by author’s group used for Natural driving data acquisition,pre-processing,fusing,storage and statistically analyzing natural driving data,it provided a data basis to establish typical driving scenario in China.(2)For the diversity and heterogeneity of natural driving data resource and the difficulty to obtain the typical possibility distribution of scenario’s constituent factors.Study proposed a method which extracts and fuses multi-dimension scenarios.This method calculates scenario’s main characters like related change of location and recognizes different situation Patrol driving,following other vehicles,Adjacent car cutin,the front car cut-out;aiming at changing lanes of driving car,study proposed a method to change lane by Dynamic Time Warping(DTW),which is calculating features of changing lanes based on time series according to distance of DTW between vehicle and lane line to distinguish scenarios that vehicles change lanes.This method calculates driving scenario’s main characters like related change of location and recognizes different situation Patrol driving,following other vehicles,Adjacent car cut-in,the front car cutout.Consequently,this approach achieved function of cut fragments automatically and extract scenarios while changing lanes in a complex situation,and provided basis of fusing multi dimension scenario.Finally,the dynamic driving scenarios are fused with other 2D main scenarios to obtain the fused muti-dimension scenarios.(3)Aiming at the imbalance between number of existing typical scenarios and dangerous scenarios,a method to generate and test the testing case from acceleration tests of simulated scenarios was proposed.A large number of typical scenarios were classified detailly by cluster analysis to design standard and typical working condition and test scenes with low complexity logically,study adopted curve fitting to calculate constituent factors and assessing parameter of scenarios under the condition of following cars,besides,the assessing method based on kernel density was implemented to acquire possibility density distribution of scenario’s parameter,the enormous amount of random test samples were generated by Metorlis Hasting sampling theorem.Then,the combining method of weighted Euler distance and PICT was applied to improve computational cost and reduce simulated computational cost based on Monte Carlo simulation.Moreover,Importance sampling mehtod was utilized to increase the possible distribution of dangerous scenes,increase the detection of tesing dangerous scenes defect of autonomous driving system,and improve efficiency of testing..(4)The most existing algorithms to generate combinatorial test scenario focused on how to reduce the size of test scenario set,but the improvement problem of contribution for each scenario to detect error,the lack of index for testing scenario’s complexity,has not been considered,which leads that distribution of scenario’s complexity generated by tested scenario sets are lack of specific description.Consequently,it is not conducive to detect defects immediately and reduce testing time.To address this problem,algorithm to generate test scenario based on combining complexity theory is proposed.algorithm implemented index of complexity to appraise effectiveness of testing scense,this method not only consided comprehensiveness for number of scense,but consided danger of testing scense,,it balanced the relationship between the number of tested scenarios and coverage.Moreover,this algorithm increased the proportion of scenarios with index of complexity in the integrated test sets,thus improves the testing efficiency.(5)Finally,study proposed a platform which can automatically generate and evaluate scenarios,automatically actuate testing process and assess results.This platform realized fully automatic execution of testing links,including rapid construction of testing scenarios,joint invocation of simulating software,result analysis and report generation.Results indicated the platform can reduce the test time of simulation with an inconceivable speed.