Routing in Scalable Delay-Tolerant Space Networks with Graph Neural Networks

Space networks are challenged by long propagation delays and prolonged link disruptions, which require Delay- Tolerant Networking (DTN) mechanisms. Contact Graph Routing (CGR) exploits the a priori available topology knowledge derived from the predictable trajectories to compute optimal end-to end routes. However, state-of-the-art CGR is limited in the scalability and stability of the computation effort, which is critical in resource-constrained spacecraft. To overcome this issue, this paper presents GAUSS: a graph neural network-based routing for scalable delay-tolerant space networks. By harvesting recent advances in Graph Neural Networks (GNNs), we can improve the scalability of CGR by a factor of two while narrowing the variability down to one-third in realistic cislunar and near-Earth systems.