**6. Summary**

Despite the advantages of automated driving technologies with respect to safety, comfort, efficiency and new forms of mobility, only driver assistance of SAE levels 0 to 2, with the first applications in SAE L3, are on the market. One of the main reasons is the lack of proof in functional safety, which is due to the immense efforts required in real world testing. Virtual testing and validation is a promising option; however, the proof of realism of the simulation is not guaranteed at the moment.

One of the main obstacles is to reproduce the performance of machine perception in the simulation. Currently, there is a huge amount of development and research ongoing in providing virtual sensors. However, there is no accepted method for the proof of realism and prognosis quality for sensor models. In the present paper, we developed a method, the *Digital Ground Truth–Sensor Model Validation* (DGT-SMV), which is based on the resimulation of actual test drives to thus allow for a direct comparison between the simulated and recorded sensor output. This approach requires defining suitable driving manoeuvres to reproduce the individual phenomena of the real sensor.

For the radar sensor, this is the multipath-propagation, separation ability and rapid fluctuation of the measured RCS over azimuth angles. The approach also requires accurate measurement equipment that records the ground truth of the driving scenario synchronously to the sensor data. After labelling the ground truth of the sensor output, a direct comparison between the simulated and recorded sensor output is possible.

For performance evaluation, we proposed a visual inspection of the simulated and recorded sensor output that we call *scatter plots* and, secondly, the transformation of these data with statistical methods based on Probability Distribution Functions to reveal the main performance of the virtual sensors. Finally, for a quick quantitative comparison, we proposed performance metrics based on the Jensen–Shannon distance. The method was applied on a commercially available sensor model (RSI radar-sensor model of IPG CarMaker) using real test drives on a closed highway in Hungary. For those tests, a high precision digital twin of the highway was available as well as the ground truth of the moving objects using RTK-GPS localization.

The results show that the DGT-SMV method is a promising solution for performance benchmarks of low-level radar-sensor models. In addition, the method can also be transferred to other active sensor principles, such as lidar and ultrasonic sensors.

**Author Contributions:** Conceptualization, Z.F.M.; methodology, Z.F.M.; software, C.W., P.L. and Z.F.M.; validation, Z.F.M. and C.W.; formal analysis, Z.F.M. and C.W.; investigation, Z.F.M. and C.W.; resources, V.R.T., A.E. and P.L.; data curation, Z.F.M. and C.W.; writing–original draft preparation, Z.F.M., C.W. and A.E.; writing–review and editing, Z.F.M., C.W. and A.E.; visualization, Z.F.M. and C.W.; supervision, A.E.; project administration, A.E. and V.R.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was not externally funded, however the preparatory work and the collection of measurement data was done in the funded Central System project, (2020-1.2.3-EUREKA-202100001) and received funding from the NRDI Fund by the National Research, Development and Innovation Office Hungary. Open Access Funding by the Graz University of Technology.

**Data Availability Statement:** Data and software used here are proprietary and cannot be released.

**Acknowledgments:** The authors would like to express their thanks to the availability of measurement data as published in [25] and those who have supported this research and to the Graz University of Technology also.

**Conflicts of Interest:** The authors declare no conflict of interest.
