On the Scaling of Reliable Interplanetary Networks With Deep Reinforcement Learning: (Invited Paper)

The rapid development of deep space exploration missions has led to the continuous expansion of interplanetary networks (IPNs) for enhanced data transfer capacities and reliability. In this paper, we propose a novel deep reinforcement learning (DRL) framework for optimizing the scaling of IPN topologies (i.e., the placement of relay satellites), such that the routing and data scheduling of interplanetary data transfer (IP-DT) in the scaled IPN achieves maximized performance gain. Our proposal leverages graph neural networks (GNNs) to extract topological correlations within an IPN and hereby learns progressive local rewriting policies for approaching the optimal solution. Extensive simulations verify the effectiveness and robustness of our proposal, showing that the learned relay satellite placements facilitate higher reliability (in terms of data delivery ratios) and lower end-to-end latency for different IPN scenarios, when compared with the existing benchmarks.