Availability and interpretability of optimal control for criticality estimation in vehicle active safety

Current and future vehicular active safety systems rely on situation interpretation algorithms for mapping sensor information of the environment to a criticality or threat value. In dangerous driving situations, these metrics are used to trigger a range of safety interventions from warning the driver, pre-tensioning of the braking system, automatic emergency braking, and automatic emergency braking and steering. For highly complex functions like automatic braking and steering, validation through real-world test drives become increasingly costly and time-consuming. Here, simulation studies can be used, where a large set of dangerous driving scenarios is labeled with a reference criticality, that should represent the true criticality. In order to find such a ground truth criticality, we propose an optimal control formulation of criticality taking into account a vehicle dynamics model as well as lane constraints. Further, using stochastically generated driving scenes, we explore the tradeoff between different goal functions and constraint formulations for interpretability and convergence of the criticality measure in simulation scenarios.