Survey Examines Federated Learning and Multi-Agent Algorithms for Satellite Constellation AI

A recent survey highlights the nascent domain of space-based artificial intelligence, pinpointing three synergistic algorithmic frameworks. Satellite constellations are transitioning from standalone units to interconnected, software-driven systems capable of real-time perception, decision-making, and adaptation in orbit. The research indicates that current AI studies primarily concentrate on individual satellite analysis, whereas autonomy at the constellation level presents entirely new challenges. These challenges encompass learning and coordination amid fluctuating inter-satellite connections, stringent SWaP-C constraints, radiation-related issues, non-IID data, concept drift, and critical safety operational limits. The three frameworks explored include federated learning for cross-satellite training, multi-agent algorithms for collaborative tasks, and collaborative sensing with distributed inference, addressing the distinct obstacles of deploying AI in space.