ESA continues to explore the value of AI in space in partnership with Thales Alenia Space and Microsoft

ESA is fostering the advent of Cognitive Cloud Computing in Space (3CS) by capitalising on high-performance Artificial Intelligence (AI) accelerator chips directly onboard satellites. In a recently agreed initiative, ESA Φ-lab will launch a challenge with Microsoft and Thales Alenia Space to develop new Machine Learning (ML) models for a hyperspectral optical sensor aboard the International Space Station (ISS).

ESA’s vision for edge computing in space is to facilitate the development of an ecosystem of in-orbit information processing, and early work done by the Agency with its partners has proved the feasibility of such an idea. In particular, the Φ-sat-1 mission experiment used a powerful onboard AI processor to successfully filter out clouds from hyperspectral optical sensor Earth observation (EO) data.

The next step was to explore the possibility of classifying additional imagery features onboard and reprogramming the chip, a breakthrough achieved on the subsequent FDL, Unibap and D-Orbit Wild Ride mission. With AI software developed by FDL, Wild Ride enabled rapid segmentation of EO data for flood identification to take place on the satellite, and the software was also shown to be reprogrammable from the ground. This year, ESA’s next-generation Φ-sat-2 satellite will deliver a platform for the in-flight uploading, deployment and updating of third-party ML models.

As part of its remit to nurture innovation in AI4EO, ESA Φ-lab has actively participated in the development of these missions and continues to provide the means for advances in in-orbit data processing. A new agreement with Thales Alenia Space and Microsoft aims to launch an open challenge for ML applications to be deployed aboard the International Space Station (ISS).