An essential component in the approval of advanced driver assistance systems (ADAS) and automated driving systems (ADS) is the quantification of residual risk, which demonstrates that hazardous behavior (HB) occurs less frequently than specified by a corresponding acceptance criterion. In the case of HB with high potential impact severity, only very low accepted frequencies of occurrence are tolerated. To avoid uncertainties due to abstractions and simplifications in simulations, the proof of the residual risk in systems such as advanced emergency braking systems (AEBS) is often partially or entirely implemented as system-level field test. However, the low rates and high confidence required, common for residual risk demonstrations, result in a significant disadvantage of these field tests: the long driving distance required. In this publication, the prediction divergence principle (PDP) is presented as an approach that has the potential to reduce the testing effort in the future, especially for systems based on the sense-plane-act structure. By continuously monitoring the prediction divergence, the approach provides essential information about the predictive performance of the system under test (SUT). In addition to the elaborated concept, this paper focuses on the mathematical decomposition of the HB into the false prediction (FPr) of the SUT and the probability that this FPr causes the HB. The approach is illustrated using the example of an AEBS. Furthermore, the prerequisites for applying the approach and the associated test reduction are derived using simplified models. Finally, the steps that must be investigated before the theoretical approach can be applied in practice are derived.
Session:
Vehicle Technology I
|
| 17:05 - 17:35