Ontology-based corner case scenario simulation for autonomous driving

Safety assessment of autonomous driving functions is an emerging topic in the automotive
industry. In order to launch an autonomous vehicle on the road, the system running
it needs to be able to react adequately in most of the cases, just like a human driver, if
not even better. Furthermore, the amount of existing data that contains corner cases is
nowhere near the amount of data needed to sufficiently train autonomous driving systems.
The identification, classification and generation of corner cases for autonomous
driving is a crucial part of scenario-based validation. In today’s world, a method that describes,
generates and classifies those at the same time, is not available.
In this thesis, a method for the description and generation of corner cases for autonomous
driving is proposed. The method uses a template ontology, as a base for the scenario
generation of corner cases with established definitions. The proposed approach also allows
combining already described scenarios into new ones without any further actions.
For an easy usage of the template ontology, an OntologyGenerator library for the simplified
generation of corner cases is provided. ... mehrThe library completely eliminates the need
for the user to work directly with the ontology and provides a method for creating scenarios
even for users with little or no experience with ontologies. Moreover, the proposed
approach is clean and understandable, making it easily scalable and extensible. The technique,
follows the structure of OpenSCENARIO, thus making it easy to use, if one already
understands OpenSCENARIO. In addition, the proposed ontology provides the base for
the classification of the described scenarios to a specific corner case category. With the
proposed technique, I was able to describe and generate seven scenarios, one for each of
the earlier mentioned corner case categories. The resulting ontologies(created using the
template ontology), that describe different corner cases, deliver correctly defined Open-
SCENARIO files, after going through the Onto2OpenSCENARIOConverter. The resulted
files deliver a correct simulation within CARLA, corresponding to the scenario description,
which shows that the proposed method functions as expected.