Data Basis for Scenario-Based Validation of HAD on Highways

Autoren

Julian Bock, MSc., Robert Krajewski, MSc., Univ.-Prof. Dr.-Ing. Lutz Eckstein, Institut für Kraftfahrzeuge (ika), RWTH Aachen University, Aachen;
Dipl.-Ing. Jens Klimke, Jan Sauerbier, MSc., Dr.-Ing. Adrian Zlocki, fka Forschungsgesellschaft Kraftfahrwesen mbH Aachen, Aachen

Zusammenfassung

Several projects about safety validation of highly automated driving systems are following a scenario-based approach. Although this approach is followed by many projects, there is no established set of scenarios. Thus, we propose a set of relevant scenarios for automated driving on highways, systematically derived based on a relational description from the perspective of the ego vehicle, and the extension of the 5-Layer model for scenario description by a 6th layer for digital information.
Furthermore, the scenario-based approach relies on measurement data from naturalistic real-world traffic events, for which a large-scale and accurate naturalistic driving trajectory dataset is still missing. The data basis is usually measured data from
driving tests, naturalistic driving studies, field tests or pilot studies. For this, test vehicles or series-production vehicles equipped with additional sensors are used. However, those measurements have several disadvantages in use for scenario-based validation. The sensors of the vehicles have limited environment perception and the scenario cannot be fully described. In addition, the recorded behavior is usually not naturalistic, since the driver is aware of the data recording while driving and the
environment is behaving differently because of the visible external sensors. Finally, this method is effort-intensive for setting up the vehicle and post-processing the measurements to get high-quality data. Within this paper, we present a method from
aerial perspective, which is strong in scenario description and naturalistic behavior, to measure data for scenario-based validation. Through the measurement from the air, a natural behavior of all road users can be ensured. Besides presenting the method, we provide a large-scale naturalistic vehicle trajectory dataset called highD-dataset recorded on German highways. Finally, we show that the dataset can be processed for scenario-based validation.

Mitglieder-Login