A Platform for Training and Evaluating Large Scale LEO Networks
Abstract
In the near future, communication networks will move into space, particularly low Earth orbit (LEO). Given the closed nature and high cost of commercial satellite deployments, it's vital that researchers have access to open platforms that allow them to develop and evaluate innovations in LEO Satellite Networks (LSNs). In this demonstration we present an ecosystem of tools developed to make both tasks easier. It includes SatGym, a flexible, framework compliant environment for training RL agents; SERENADE, a large-scale, realistic satellite network emulator; and SatGym-Server, an HPC bridge application that enables LSN controllers (including RL agents) to interact with SERENADE in a fully distributed way. Together these tools allow researchers to develop and evaluate large-scale, decentralized LSN orchestration controllers in real time with realistic traffic generation and orbital dynamics.