Research On Key Scenarios Of Lane Changing For Autonomous Vehicles Based On Functional Testing

Test scenarios play an important role in the development of autonomous vehicles.The completeness of the test scenarios helps to verify the capabilities of autonomous vehicles’ perception and decision control.However,in the natural driving scene library,non-safety-critical scenes dominate and dangerous-critical scenes account for a small proportion,which results in a limited number of dangerous-critical scenes that can be selected in the test.This makes it impossible to fully verify the driving safety of autonomous vehicles.This paper studies the generation methods of key test scenarios for autonomous vehicles.The main research contents are as follows:(1)Based on the function test conception,a method of applying fault tree analysis to determine the constituent elements of the test scenario is proposed.Based on the position and movement direction of the obstructive vehicles around the main vehicle,according to the principle of "from less to more" for the number of lanes,and " replace complexity with simplicity" for the number of obstructive vehicles,and using the method of pairwise combination to design a combination group of test scenarios with different complexity.A functional scene library including lane keeping behavior,lane changing behavior and lateral following deviation behavior of the main vehicle is generated on the two-lane road segment and the three-lane road segment.(2)Based on the model predictive control,selecting the vertical and horizontal decoupling motion planning and control framework,and designing a lane changing control algorithm.Which enables the main vehicle to change lanes autonomously according to the status of the obstructive vehicles.The algorithm considers three situations that occur when the main vehicle changes lanes,and enables the main vehicle to perform local dynamic programming in real time during the lane change process.Under the joint simulation platform of Matlab / Simulink and Carsim,the main vehicle was changed at different speeds under three operating conditions,and the lane change control algorithm was verified to meet the real-time and comfort requirements.The algorithm laid a good algorithm simulation foundation for the generation of logic scene library in the following chapters.(3)Based on the minimum safety distance model of lane change,the parameters of the lane change scene are designed.The parameters are combined using a pairwise combination method to obtain 376 sets of basic dynamic scene parameters.Based on simulation,the basic dynamic scene parameters are screened by using 4 safety indexes and 2 comfort indexes,and a key logical scene with 166 sets of dynamic scene parameters is obtained.Dangerous critical scenes that do not collide account for 22% of the total number of scene groups.Compared with only 2% of dangerous driving scenes in natural driving scene data,the number of dangerous critical scenes is greatly increased.166 groups of dynamic scene parameters were clustered by weighted Euclidean distance,and 31 groups of representative dynamic scene parameters were obtained.Finally,based on the task form combination method,specific key scenarios containing static elements and dynamic elements under the main vehicle lane changing function on a two-lane road segment are designed.