Evaluating the Realism of Lateral Movement Modeling With and Without the Consideration of Influencing Factors

The lateral offset between vehicles is a relevant factor for the range of vision of vehicle sensors and thus influences the output of an automated driving function relying on the sensor input. With simulations being an integral part of the automated driving function validation process, their ability to realistically reflect the lateral movement of vehicles is essential. A model addressing this task has been introduced in earlier work. How to assess the realism of the generated lateral movement is a question of its own. Earlier work bases the evaluation on comparing the distributions of selected features for the real and artificial lateral movement using the Jaccard index. When using this unit as a basis, the model generates not only qualitatively but also quantitatively realistic results and, in particular, outperforms other existing approaches. To gain more insights on the model’s abilities and limitations, we explore the use of the Cramér-von Mises distance and the associated test as further means for evaluating the baseline model and, in particular, the enhanced version of the model considering influencing factors. The results show that the Cramér-von Mises distance and test are suitable additional means for comparing the feature and lateral position distributions, underlining the strengths of the model but also revealing its weaknesses. The introduced factor-based Cramér-von Mises distance is in particular able to bring out the effect of considering influencing factors on the model results.