Constellation Autonomy: AI solutions for adaptable and efficient operations
- Paper ID
86874
- DOI
- author
- company
European Space Agency (ESA-ESOC); ESA - European Space Agency; Saarland University; INRIA; GMV GmbH; GMV Innovating Solutions; ; GMV Insyen AG; Thales Alenia Space France; Thales Alenia Space
- country
Germany
- year
2024
- abstract
The space industry is witnessing a surge in satellite constellation launches, a trend set to intensify. Constellation missions offer sophisticated solutions to complex problems, often employing hundreds of collaborative spacecraft, to provide enhanced services in Positioning, Telecommunications, and Earth Observation. They ensure continuous global coverage, frequent interaction, and improved network redundancy and capacity. "New Space" characterizes ventures aiming to democratize space through smaller, cost-effective spacecraft integrated into larger systems.\\ The constellations of assets is diversifying as the number of satellites in Low Earth Orbit (LEO), Medium Earth Orbit (MEO), and Geostationary Earth Orbit (GEO), as well as stratospheric platforms, continues to rise. This expansion leads to the formation of a heterogeneous network, facilitating enhanced connectivity among these assets. Earth Observation and Telecommunication systems are growing in complexity, involving various interconnected space and airborne assets with demanding customer requirements. Managing these complex systems requires advanced computational methods and continuous situation-aware replanning, extending the existing Concepts of Operations.\\ The growing complexity of managing satellite constellations prompts the use of more advanced autonomous ground systems for planning optimisation. Artificial Intelligence (AI) can be crucial for addressing the exponential increase in tasks associated with a rising number of assets in constellations. AI is seen as a game changer, offering more flexibility or even surpassing traditional optimization techniques for automating and optimizing the network management of satellite fleets. This paper will present the work done in an industry-driven activity, implemented by the European Space Operations Centre, with a Consortium of GMV, the Saarland University, Thales Alenia Space France and major operators as external consultants. The paper will include the identification, thorough analysis and prioritisation of numerous use cases where AI can support constellation missions’ management, the resulting AI-based Concepts of Operations, and the provision of solutions for two realistic resource planning scenarios, applying a prominent AI method, Reinforcement Learning (RL). The work explores the optimization of battery and memory resource utilization, alongside the enhancement of multi-hop routing within mega-constellations. Leveraging RL techniques, we train agents within intricately detailed and sophisticated environmental models, comparing to and surpassing the limits of classical optimization methods. This approach facilitates a scalable, flexible, and generalizable decision-making process, crucial for efficient mission operations in dynamic and complex mega-constellation environments.