LEO Satellite Networks

Low Earth Orbit satellite networks study how constellations of fast-moving satellites can route data, hand off users, schedule observations, and share limited power, storage, and communication links. Common approaches include graph neural networks and reinforcement learning for routing and handover, optimization methods for task scheduling and topology, and federated learning for training models across satellites without moving all data to Earth. Researchers evaluate these methods in constellation simulations, satellite internet measurements, remote-sensing benchmarks, and, increasingly, on embedded hardware. A central challenge is making learned systems reliable under changing connectivity and uneven data while meeting tight onboard resource and latency limits.