Opportunities and challenges of AI on satellite processing units

Abstract

Higher autonomy in satellite operation is seen as the key game changer for the space systems market in the next decade, with a considerable amount of agencies and startups focusing on bringing machine learning to space. The adoption of Artificial Intelligence on-board of satellites is still limited due to the processing capabilities of radiation hardened hardware, which requires flight-heritage and extensive qualification. At the same time, the satellite market is undergoing a major paradigm shift from a hardware equipment perspective. Classical approaches, which aim at realizing satellites compliant with mission profiles including a long-lasting operational life and an extremely high reliability are ill-suited for many of the new market segments. The satellite-manufacturing industry is gradually adapting to these new mission requirements by identifying segments where components-off-the-shelf (COTS) can be employed. The latest generation of commercial components offer the unique possibility to integrate AI-algorithms with relative ease with tool assisted design and a much higher performance in parallel processing. In this position paper, the authors introduce the state of art of on-board AI and present the approach that is currently being researched in Airbus Defence and Space to perform neural network inference in various mission scenarios.