The development and test of ADAS and Automated Driving Systems (ADS) require appropriate scenario data. To ensure the correct functionality and functional safety of such systems, an incredible amount of scenarios is necessary, containing normal, critical, and accident situations. These scenarios are usually used for virtual simulations. However, selected scenarios should be also physically tested on proving grounds. We developed a method to extract and describe maneuver-based and parameterized scenario catalogues for development and test of ADAS and ADS.
We used real accident data from GIDAS (German In-Depth Accident Study). The focus was on car accidents in urban areas as the complexity in urban traffic is much higher than on highways (heterogeneous infrastructure, large variety of road users and behavior).
At first, we clustered the (weighted) GIDAS accidents into different scenario groups. Then, we identified relevant parameters that are necessary for the description of the static and dynamic content of scenarios. The static content was extracted within the “environment analysis”. With this, the scenarios can be parameterized in terms of weather and lighting conditions, road layout (e.g. number of lanes, road width etc.).
For the “dynamic analysis” we additionally used the GIDAS-PCM, containing reconstructed maneuvers, time- and location-resolved trajectories, accident sequences. Here, we generated statistical descriptions about speeds, trajectories, braking or steering maneuvers. Finally, some concrete example scenarios have been transferred to IPG CarMaker and OpenDRIVE / OpenSCE-NARIO files.
With the developed method it is possible to transfer thousands of single traffic events and/or accidents with concrete characteristics into generic (test) scenarios. Within the project, scenario groups have been created using a maneuverbased approach. There are currently four main categories (following in one lane, crossing scenarios, turning scenari- os, and lane change) which are further divided into sub-maneuver groups.
The created parameter sets per scenario group contain several static and dynamic parameters. These distributions can be used by system engineers for virtual simulation runs (e.g. with randomly varied scenarios) but also by test engineers to parameterize physical tests. The approach was already tested with partners with demonstrations in physical tests.
The implementation in concrete formats (IPG CarMaker, OpenX) showed that an automated transfer is not possible at the moment due to the complexity and multitude of implementation options. The developed method works for accident data out of GIDAS and was already tested in physical tests. However, the method was not yet applied to normal/critical situations but this should also work with the presented static and dynamic parameter sets. Another limitation is the lack of automatic data transfer from the PCM format into the open ASAM standards (OpenX).
As scenario catalogues are essential for virtual simulations as well as for physical tests of ADAS and AD functions the presented method helps to provide appropriate scenario data out of real-world accidents. The big advantage is that the created parameter sets and scenarios base on reconstructed accident data and can be used independently from certain software solution or format.