Machine Learning Based Testing Scenario Space and Its Safety Boundary Evaluation for Automated Vehicles

Abstract

Autonomous vehicles (AVs) must be thoroughly evaluated before their release and deployment, so testing scenarios in simulations and closed facility tests are very important. The research presented in this paper focuses on the classification of scenario elements and the safety boundary of the scenario space. Scenario elements are divided into the static scenario elements and the dynamic scenario elements, and both of the two parts can be further divided into two types: the elements outside or inside the road. The importance weights of scenario elements are determined by Analytic Hierarchy Process (AHP) method and the effect transmission model to construct a scenario space. The relative speed, Time To Collision (TTC) and Time Headway (THW) are selected as risk assessment parameters. Support Vector Machine (SVM) is used to solve the safety boundary of the scenario space. Test results show that the classification of scenario elements is reasonable and the importance weights of scenario elements provide a basis for constructing the scenario space, and the regression lines of the safety boundary can distinguish dangerous driving states from safe ones.