Automated test case generation for satellite on-board image processing using reinforcement learning

Kurzfassung

In the satellite domain, on-board image processing algorithms must run and must be maintainable in space for several years. To ensure that all requirements are met, the algorithms must be thoroughly tested. This is a challenge due to the large input range of the algorithms, as it is not always possible to test all possible inputs. Therefore, there are approaches using e.g. genetic algorithms to generate inputs that provoke critical behavior. However, such algorithms have some problems, which is why we propose a reinforcement learning approach to automated generate test cases for on-board image processing algorithms. In this thesis, we evaluate our approach based on the PLATO fine guidance system algorithm as a case study.