Wang, YaYaWangGrabowski, MaximilianMaximilianGrabowskiPaschke, AdrianAdrianPaschke2023-06-192023-06-192022https://publica.fraunhofer.de/handle/publica/443016In recent years more and more deep learning-based models have been applied to several tasks in autonomous driving and achieved encouraging results, such as object detection, traffic scene segmentation, and path planning. However, those models are often not interpretable and prone to fail in some edge cases. Especially for the case of rule exception, autonomous driving systems are required to fully understand traffic situations, and apply specific rules that are not explicitly stated in current traffic regulations, for proper decision making. The ability to solve these cases is difficult to learn inductively from statistics without using world and normative knowledge. In this paper, we demonstrate that a transparent ontology-based model can assist vehicles in resolving exceptional cases to comply with traffic regulations by reasoning over perceived data combined with formalized traffic rules.enautonomous drivingrule exceptionknowledge representationconference paper