Research And Implementation Of Virtual Road Network Environment Construction And Scena Rio Automatic Generation Algorithm For Autonomous Driving Simulation Testing

With the rapid development and application of autonomous vehicle,people are concerned about the safety while expecting the convenience it brings.Autonomous vehicles need to be comprehensively tested to verify their safety before large-scale commercial use.Current testing methods involve public road testing,proving ground testing,and virtual simulation testing.Public road testing suffers from its insufficiency of scenario repeatability,high cost and risky.Proving ground testing has been beset complexity on building test scenarios and low scenario authenticity.Virtual simulation testing can solve these problems.The virtual simulation testing requires the construction of realistic and typical test scenarios.In this regard,it is necessary to design a typical static road environment and provide realistic motion trajectories for dynamic traffic participants in the test scenarios.To address the above problems,this thesis proposed a static road environment construction method based on genetic algorithms using real map data to generate a set of typical test roads,forming a test road network for test scenarios.Based on real vehicle operation data,a lane change scenario generation model was designed using generative adversarial network to realize automatic generation of lane change test scenarios.Virtual simulation test scenarios were built in Pre Scan to realize the simulation test of autonomous vehicles.The main research content of thesis contains the following three aspects:(1)Proposing a road environment construction method for autonomous driving test scenarios using real road network data.Through analyzing real road data in the target area,the road unit models have been established based on the geometric characteristics of the road.Typical road units and their parameters have been obtained through cluster analysis on different classifications of real road network data.The genetic algorithm is used to optimize the combination of different typical road units in a limited space to form a typical road network and achieve the coverage of the real road environment in the target area.(2)Constructing a W-DCGAN model based on real trajectory image data is constructed.In this thesis the lane change cut scenarios were extracted from high D dataset and classified by the hazard level based on TTC model.Combining the design ideas of CGAN,DCGAN and WGAN,the W-DCGAN model based on real trajectory image data is designed to realize the automatic generation of lane change trajectories for high-risk scenes.(3)The generated road network and lane change trajectories were combined to establish a test scenario in Pre Scan.Firstly,the road network environment of the test scenarios was built on Pre Scan,Secondly,the AEB control algorithm based on the TTC model was designed using Pre Scan and Simulink jointly.The automatically generated vehicle lane change trajectories were imported in virtual test scenarios to set up four kinds of test scenarios according to different danger levels.Finally the virtual simulation test of autonomous vehicles is executed in test scenarios.The test road environment construction method proposed in thesis can provide a typical road network for test scenario construction based on real road network data in the target area.The test scenario generation method based on generative adversarial network can realize test scenario generation with different hazard levels.The simulation tests show that the road network environment and the lane change trajectory generation method proposed in thesis can support the virtual simulation test of autonomous vehicles.The methods proposed in thesis can provide guidance for the construction of test scenarios and promote the application of autonomous vehicle technology.