9 in 10 cameras agree: Pedestrians in front possibly endangered

Modern cyber-physical systems integrate data from many sensors as a regular part of their operations. Over the years, researchers have proposed methods ranging from statistical approaches to neural networks to achieve this sensor fusion, along with high-level paradigms such as early and late fusion. However, quality assurance of sensor fusion algorithms typically focuses on highlighting their accuracy or ability to reduce uncertainty under given conditions. This paper aims to establish a qualitative approach to testing sensor fusion. We formulate an answer set program based on desirable properties for sensor fusion and show how to apply it to test fusion algorithms. Our results indicate that our approach is effective at finding faults, but does not easily find minimal models for large inputs.