Evolutionary generation of test scenarios for autonomous driving

Course information

  • Practical course (Praktikum) for Master students
  • Winter semester 2016/17
  • Lecturer: Florian Hauer

Description

The complexity of modern driver assistance systems increases continuously. This makes testing them more and more difficult. Significant problems are caused by the large input space and complex interactions with the environment.

On the one hand, human testers cannot easily identify all relevant test scenarios (they can be non-intuitive), and on the other hand, the number of possible test scenarios is very large so that they cannot all be used. A possible solution is to generate and select test scenarios algorithmically.

In this practical course, we want to implement and evaluate an evolutionary approach [1] using the example of a parking assistant. For this, we will first implement abstract models of the parking assistant and the environment and then experimentally evaluate the possibilities and limitations of the approach.

We will use MATLAB, since MATLAB provides according libraries and visualization capabilities.

This practical course offers the opportunity to learn about model-based testing and evolutionary algorithms. Additionally, you can deepen your knowledge about MATLAB which is of high importance in the fields of scientific computing, machine learning, avionics, and the automotive industry.

[1] Buehler, Oliver, and Joachim Wegener. “Evolutionary functional testing of an automated parking system.” Proceedings of the International Conference on Computer, Communication and Control Technologies (CCCT'03) and the 9th. International Conference on Information Systems Analysis and Synthesis (ISAS'03), Florida, USA. 2003.

Prerequisites

Working knowledge in MATLAB.

Detailed information

See Campus for more information.