Automatic Disengagement Scenario Reconstruction Based on Urban Test Drives of Automated Vehicles

Abstract

In recent years, scenario-based testing has gained increased attention as a potentially efficient strategy for validating the overall safety of automated vehicles. However, which scenarios are of interest for testing and how to systematically generate the test instances remain as unanswered questions. In this work, we interpret the importance of incorporating automated vehicle disengagement scenarios into scenario-based testing. Accordingly, we design and implement a fully automatic pipeline to reconstruct the essential and error-reduced disengagement scenarios based on imperfect perception measurement data from real test drives in an urban environment. Our concept is developed based on 137 disengagement data snippets and two additional datasets for handling false positives and false negatives in the original measurements. We use additional disengagement snippets for validating the performance of the pipeline. We exhibit representative reconstructed scenarios to show a successful restoration of the reality and quantitatively demonstrate the correct functioning of the methods in the pipeline regarding filtering irrelevant objects and handling the perception inaccuracies.