Wednesday, May 31, 2023
Machine learning-based Generation of realistic Edge-Case traffic scenarios
Topic and Goal of the Thesis
How can we test, if automated driving is safe?
This is a central question in research in safety assurance for automated vehicles. A central approach for this is scenario-based testing. Particularly exciting are edge cases that challenge the system but are still intended to be realistic. However, such edge cases rarely occur in reality and can hardly be found in data.
In this thesis, a method for generating realistic edge cases of driving scenarios based on real data shall be developed. In particular, it shall be analyzed how the behaviour of dynamic road users can be without creating driving scenarios that never occur in traffic. Both, machine learning and rule-based approaches can be used for this purpose.
Working Points
- Literature research on the topics of data classification and Edge-Cases in the context of driving scenarios
- Development of a methodology to classify and generate realistic Edge-Cases for dynamic road users
- Testing of generated scenarios in simulation
- Validation of the methodology based on real intersection data
Requirements
- Experience with python
- Good English or German language skills
- Reliability, commitment and enjoyment of working independently as well as methodically
- Enthusiasm for automated driving
Note: Please attach brief resume and grade summary.
Contact
Christoph Glasmacher M. Sc.
+49 241 80 25611
Email
Type of work
Masterarbeit
Start
Earliest possible date
Prior knowledge
Python
Language
Deutsch, Englisch
Research area
Fahrzeugintelligenz & Automatisiertes Fahren