Towards Criticality Characterization of Situational Space

The assurance of safety is crucial for the introduction of automated driving. Today's testing relies on expert knowledge of critical situations. Real-world and simulation tests are carried out to cover as many test cases as possible. The more extensive the tests, the safer the automated function is assumed, but it is uncertain if all critical situations are covered. Therefore, a mathematical generalization to represent the situations is introduced, the situational space. In order to support the selection of situations to be tested, a procedure to examine the situational space is presented in this paper. Therefore, necessary definitions are provided and the used methods are explained. Additionally, the required datasets are generated on simulated data and classified with support vector machines. Thereby, a characterization of the situational space is achieved, which is the main contribution of this work. Furthermore, the results are compared to and evaluated on real-world situations, that were extracted from recorded test drives in mostly urban traffic.