Testing automated and autonomous driving systems is challenging for various reasons, e.g., because of the absence of clear and precise specifications of the system under testing, because of the unclear test selection criteria for “good” test cases, and because of missing test exit criteria. As a consequence, a popular approach suggests reusing recorded test drives as test cases regardless of the system being tested. We show that this is generally equivalent to purely random testing of the system, that a recorded test drive may be sometimes “good” and sometimes irrelevant, and that it is necessary to derive new test cases for each new system. A solution to this issue is a simulation-based heu… Mehr anzeigen