Presentation Description

Safety assurance for automated driving is one of the major challenges in automotive research. Scenario-based testing has become a promising approach to tackle this issue to assess the risk of these functions. In this approach, the automated driving function is confronted with clearly defined scenarios instead of driving in real-world traffic. Despite its potential of significantly reducing the required testing effort compared to real-world drives, this approach comes with three major challenges: The proper definition of scenarios covering real-world traffic sufficiently and serving the safety assurance process, the management of those scenarios and underlying data, and the acquisition of traffic data to link scenarios to the real world, see PEGASUS project [MAZ19]. The poster illustrates an approach for collection of large amounts of data for mining of relevant scenarios. The data consists of precise motion data of road users collected with a UAV from a bird’s eye view. The poster outlines the challenges, solutions, and possibilities of this scenario mining concept.
In addition to the information provided by the object instance and motion information, digital maps such as Lanelet2 and ASAM OpenDRIVE expand the benefit of the data. This enables the creation of semantic relations between road objects and infrastructure, which forms the foundation for scenario extraction. These datasets can be created on virtually any traffic domain e.g. on highways, as shown in highD [KRA18], in urban areas like in inD [BOC20], at roundabouts, parking lots, truck depot or test tracks. This allows to collect relevant data from a wide range of traffic domains, which supports the growing demand for all kinds of scenarios.
levelXData holds a large database filled with various types of trajectory data from all over the world [LXD24]. A sophisticated scenario extraction process allows to filter for specific scenarios considering different traffic structures. The modular process allows for customization, enabling the extraction of scenarios to fit different concepts. As a result, it is possible to consistently generate scenarios, which align with a given scenario model. This approach offers a practical way to analyze movement patterns in various environments.
The poster showcases the extraction approach for large-scale trajectory datasets based on an example for highway scenarios. Furthermore, examples for relevant statistical analyses and distributions of scenario parameters are presented.

Speaker Biography

Presentation File

View Presentation