Design And Implementation Of Critical Scenarios Generation Tool For Autonomous Driving Systems Simulation Test

The safety issues of autonomous driving systems have been widely concerned for a long time.Self-driving cars must be thoroughly tested before they can be driven on real-world roads.Traditional vehicle road tests are expensive and difficult to conduct relatively complete tests on the safety of autonomous driving systems.To address these challenges,industry and academia have proposed simulation testing methods based on virtual scenario which expose the defects and deficiencies of autonomous driving systems by searching for challenging scenarios(ie,critical scenarios).The traditional critical scenario generation technology based on the search method has the following problems:(1)the generated critical scenarios have a high similarity,and fewer safety violations of the automatic driving system can be found;(2)during the testing process,multiple,Repeatedly connecting the automatic driving system and resetting the virtual scenarios of the simulator causes a lot of unnecessary time consumption in the simulation test process,which affects the testing efficiency of the automatic driving system.In order to solve the above problems,this thesis proposes a critical scenarios generation method based on evolutionary search,which uses a multi-objective genetic algorithm to optimize the search direction of participants’ behaviors and induces the autonomous driving system to perform driving actions that violate safety norms.In order to more effectively expose the safety violations of the autonomous driving system,composite behaviors are introduced into the method to improve the interference of NPC vehicles on the autonomous vehicle(i.e.the ego vehicle)in the scenarios;in order to discover more types of security violations,A multi-objective fitness function is defined to select scenarios with high risk potential and guide the crossover and mutation between scenarios,which reduces the similarity of generated scenarios while increasing the challenge of scenarios.In order to improve the test efficiency,a spatiotemporal continuous test execution method is designed,and the scenario is generated by executing the search process in parallel;during the test process,the coordinates of the scenario initialization are transformed to realize the continuous execution between different scenarios and reduce the reconnection of the automatic driving system.Number of resets with the emulator.Finally,based on the research results of the above key technologies,a critical scenario generation tool for automated driving simulation testing is designed and implemented.Experimental results show that,compared with existing methods and tools,the proposed method generates more safety violation scenarios in the same time,and discovers more types of safety violations in autonomous driving systems.