Data-Driven Scenario Analysis for the Validation of Automated Driving Functions

The validation of automated driving functions increasingly relies on scenario-based testing. A key challenge is the systematic identification, structuring, and analysis of traffic scenarios. The objective of this project is to develop data-driven methods for the automatic extraction, clustering, and evaluation of traffic scenarios from real-world traffic data.

The focus lies on a novel approach in which scenarios are defined based on the observable behavior of a vehicle. Changes in driving behavior indicate the beginning and end of individual scenarios. These scenarios are subsequently clustered using modern machine learning techniques to identify structural similarities and derive representative scenarios.

Another key aspect is the evaluation of scenario coverage. This involves analyzing to what extent the identified scenarios (or scenario clusters) represent real-world traffic situations and assessing the uncertainties arising from previously unobserved scenarios.