Research On Key Test Scenarios Of Autonomous Vehicles At Crossroads Based On Simulation

Self-driving car testing is an important part of the development process of self-driving cars.Although the natural driving scene library used by enterprises for testing can reflect the real traffic situation,the non-critical scenes in the natural driving scene library dominate,and the key dangers The proportion of scenes is very small,which leads to a limited number of key scenes to choose from in the test,and it is impossible to fully verify the driving safety of autonomous vehicles.This paper studies the method of generating key test scenarios for autonomous vehicles.The main research contents are as follows:(1)A method for generating basic test scenarios for autonomous vehicles is proposed.Taking a one-way two-lane intersection without traffic lights as an example,based on the most complex vehicle position combination group under this working condition,according to the driving direction of the main vehicle,a full combination strategy is used to combine the movement direction of the interfering vehicle,and screen according to the designed scene The scenarios are screened in principle,and the basic test scenarios are obtained.(2)Analyze the complexity of basic test scenarios based on analytic hierarchy process and graph entropy.First,the analytic hierarchy process is used to calculate the weight and total value of each scene element class,and then the complexity calculation model of the traffic scene is established based on the graph entropy,and the basic test scene of the main vehicle going straight at the intersection is taken as an example to calculate the complexity.(3)Design motion planning and control algorithms based on model predictive control.A dynamic safety zone-based obstacle avoidance strategy is proposed for crossroads.This strategy combines horizontal safety to plan a dynamic safety zone for autonomous vehicles under the premise of first ensuring vertical safety.Then,based on the multi-constraint LTV-MPC,the strategy uses the inclusion of linear The vehicle dynamics model of the tire model is designed with an integrated algorithm for the motion planning and control of autonomous vehicles,and the effectiveness of the algorithm is verified on the Matlab/Simulink and Carsim co-simulation platform.(4)A simulation-based generation framework for key scenes at intersections is proposed.The dynamic scene parameters are combined based on the three-parameter combination strategy.Through scene simulation,the dynamic scene parameters are screened using 3 safety indicators—collision time,post-encroachment time and the absolute value of longitudinal acceleration,and 617 groups of two interfering vehicles participating are obtained.Key scenes and 352 groups of key scenes involving a jamming car.Two key hazard scenes involving jamming vehicles accounted for 20.1%of the total number of scenes,and one key hazard scene involving jamming cars accounted for 18.5% of the total number of scenes,which is similar to natural driving.Compared with the occurrence rate of dangerous scenes in the scene is only 2%,the occurrence probability of key dangerous scenes is greatly increased.Finally,the above key scenarios are clustered using weighted Euclidean distance based on information entropy,and 9 groups of key test scenarios involving two jamming vehicles and 7groups of key test scenarios involving one jamming vehicle are obtained,which can be used for closed field safety testing.