Dr. Rainer Rasche
Group Leader Software Development
dSPACE GmbH

Highly Automated Driving (HAD) is one of the big challenges in current automotive development, making it a transition phase between existing driver assistance systems and upcoming fully autonomous driving in the future. Objectives such as keeping safety distances, optimal energy consumption or passenger comfort levels need to be tested for various parameters or traffic scenarios. As a resulting question, a responsible engineer may ask: “How do I drive the correct million virtual test kilometers over night?”

Classical test methods do not seem to be sufficient, as they rely on existing test catalogues. Based on these, we need to generate vast amounts of test variants, but in a smart way, to (a) quickly reveal violations of multiple acceptance criteria, and (b) guarantee an adequate coverage of relevant traffic situations. Stochastic evolutionary algorithms, based on test result histories, generate tests towards critical situations.

To handle the resulting high test load, the use of concurrent cluster Simulation setups is mandatory. Access to test and result management enables the evolutionary “intelligence” of test generation, while standardized interfaces enable easy integration of these algorithms and keep the tool chain modular and adaptable. In the end, this enables an engineer to use his local tool chain to debug and inspect failed test variants easily.