Observation Based Creation of Minimal Test Suites for Autonomous Vehicles

Abstract

Autonomous vehicles pose new challenges to their testing, which is required for safety certification. While Autonomous vehicles will use training sets as specification for machine learning algorithms, traditional validation depends on the system's requirements and design. The presented approach uses training sets which are observations of traffic situations as system specification. It aims at deriving test-cases which incorporate the continuous behavior of other traffic participants. Hence, relevant scenarios are mined by analyzing and categorizing behaviors. By using abstract descriptions of the behaviors we discuss how test-cases can be compared to each other, so that similar test-cases are avoided in the test-suite. We demonstrate our approach using a combination of an overtake assistant and an adaptive cruise control.