In order to significantly reduce the testing effort of autonomous vehicles, simulation-based testing in combination with a scenario-based approach is a major part of the overall test concept. But, for sophisticated simulations, all applied models have to be validated beforehand, which is the focus of this paper.
The presented validation methodology for sensor system simulation is based on a state-of-the-art analysis and the derived necessary improvements. The lack of experience in formulating requirements and providing adequate metrics for their usage in sensor model validation, in contrast to e.g. vehicle dynamics simulation, is addressed. Additionally, the importance of valid measurement and reference data is pointed out and especially the challenges of repeatability and reproducibility of trajectories and measurements of perception sensors in dynamic multi-object scenarios are shown. The process to find relevant scenarios and the resulting parameter space to be examined is described. At the example of lidar point clouds, the derivation of metrics with respect to the requirements is explained and exemplary evaluation results are summarized. Based on this, extensions to the state-of-the-art model validation method are provided.
