Verification and Validation Methods for Automated Vehicles of Level 4 and 5

The goal of VVM is to develop, prototype and implement application-oriented concepts for a complete, consistent safety demonstration for highly automated driving with a minimum of necessary test scope and test effort.

To this end, the most important German OEMs, Tier 1 suppliers and research institutes are working together to meet the challenge of safety verification. The Institute of Automotive Engineering (FZD) is mainly responsible for the investigation of different approaches for data reduction, as well as requirements definition and implementation for sensor models in urban environments.

Data reduction

When automated vehicles are operated in the field, the data coming from the sensors is too large to be recorded. In order to detect and store only scenarios relevant for safety verification of highly automated driving, data reduction is essential. FZD investigates in particular the following two approaches to data reduction in VVM.

  • Comparison of human driving with an automated driving function that runs parallel to human driving to identify differences in trajectory selection.
  • The detection of “surprises” in system behavior, based on a comparison of available sensor data to extended ground truth information, e.g. by retrospectively assessing prediction models.

In the first phase of the project, approaches to data reduction will be explored and necessary algorithms will be developed accordingly. In the second phase, the proposed approaches will be implemented and tested in a test vehicle.

Sensor modelling

As the VVM project focuses on urban areas, which are different from the use cases from PEGASUS and ENABLE-S3, the requirements for active environmental sensor models in VVM will be re-analysed. Additionally, the sensor models developed in SET Level 4 to 5 will be implemented and verified in this urban environment.