Improving scenario selection for simulations by run-time control-flow analysis

Cyber-physical systems like active safety systems in recent vehicles are significantly driven by software and rely predominantly on data that is perceived by cameras, laser scanners, and the like from the system's environment. For example, these sensor-based systems realize pedestrian protection function-alities, which cannot be tested under simplified conditions on proving grounds only or by arbitrary test-runs on public roads anymore. Instead, simulative environments are used nowadays, which provide the virtual surroundings for such a system where its real input sources are replaced with simplified sensor models. Thus, interactive and hazard-free system tests and automated system evaluations can be carried out easily. However, the simple strategy to run all available modeled traffic scenarios in the simulation on any change of the implementation would consume too much computation time to provide effective and fast feedback for developers. In this article, an improved strategy for selecting scenarios that shall be run in a simulation based on run-time control-flow analysis is proposed, which resulted from the in-depth analysis of the revision history of the source code and their accompanying simulations for two self-driving vehicles. The outlined strategy is evaluated on a self-driving miniature vehicle.