Multi-Agent DRL for Distributed Routing and Data Scheduling in Interplanetary Networks

With the fast development of deep space exploration missions, the data transfer in interplanetary networks (IPNs) is gaining increasing attention. In this work, we propose a deep reinforcement learning (DRL) based routing and data scheduling approach, which leverages a multi-agent setup for distributed operations and aims to balance the trade-off between average end-to-end (E2E) latency and delivery ratio of interplanetary data transfers (IP-DTs) well. Specifically, DRL agents based on asynchronous advantage actor-critic (A3C) are deployed on each IPN node to handle the routing and data scheduling of IP-DTs there separately. Simulation results confirm that our proposal can handle the routing and data scheduling of IP-DTs more adaptively and balance the tradeoff between the delivery ratio and average E2E latency better than the benchmarks.