The validation of autonomous driving functions need driving scenarios
Evaluating the safety of driver assistance systems and autonomous driving functions is extremely complex and time-consuming. The direct interaction of vehicles with other road users creates an enormous amount of possible driving situations that need to be validated and verified. The efficient management of the underlaying driving scenarios is crucial for this task.
Driving scenarios are either recorded in test vehicles with various sensors or generated on the basis of so-called scenario definition languages (SDLs), such as ASAM openSCENARIO. Under certain circumstances, several terabytes of scenario data are created every day. A central data repository can ensure that different user groups have access to the existing scenarios in order to use them for a variety of validation purposes.
openMDM is a suitable platform that can be used to implement such a data repository. The underlying ASAM ODS standard provides the necessary flexibility for data storage.
An application scenario could look like this
First, the original file formats of the recorded and generated scenario data are imported, classified in terms of content and systematically described with metadata. Standardized labeling methods such as ASAM openLABEL can support this processing step. The metadata is transferred to freely definable templates that can be composed of several components. This allows, for example, to comprehensively document the “maneuvers”, “objects” and “actions” contained in a scenario. When importing, the individual scenario files are linked to the meta data templates which describe them.
Based on the meta data, the repository users can search for specific scenarios. Both a search based on individual attributes and a full text search are possible. For example, the user can thus find scenarios that represent a “threading maneuver” with “3 vehicles” on a “three-lane highway”.
With the found scenarios, the user can define test cases, including the description of the operation domain (ODD) and the device under Test (DuT). In the next step, via export and download mechanisms or direct data access, the defined test cases can be made available in a suitable format for simulations and road tests.
The results of the executed validations can then be imported again and assigned to the test cases. In this way, a comprehensive test case catalog for the validation of driver assistance systems and autonomous driving functions is created over time.
Added value through standardization and openness
A key feature of this solution – compared to proprietary data management solutions – is the open REST interface. Different client applications can use it to access the repository in a standardized way. At the same time, the web-based software architecture of openMDM ensures that the application is scalable in terms of data volume, number of users and distributed usage. As a result, the solution can also be used in a global environment for the cross-departmental and cross-company management of large quantities of driving scenarios, test cases and validation results.