Lunar Network Autonomous Routing (LEARN)

Lunar communication architecture comprises of various types of connections such as Earth to Satellite links, Lunar Orbit and Surface networks, etc. Lunar networks have multiple agents with different specifications, requirements, and interactions. These heterogeneous networks introduce many deficient factors including unscheduled and unpredictable behaviors, intermittence and interference, high latency tolerance data. Therefore, a novel, LCA-native network management solution needs to effectively address these variables. In this work, we propose to develop, emulate, test, verify, and validate an innovative novel Artificial Intelligence (AI)-based Lunar nEtwork Autonomous RoutiNg (LEARN) technology. LEARN gives an autonomous lunar network management solution in two main categories; (i) optimized various data routing (ii) intelligent packet scheduling and resource control to select data packets, traffic data flow, route, resource and communication modes. We consider the delay-tolerant networking (DTN) architecture and protocol to maintain some of the issues and our LEARN technology will bring intelligent network routing and scheduling beyond DTN. To this end, we adopt a systematic and modular approach to build a Multiple Agent Deep Reinforcement Learning (MADRL) framework for LEARN. We have proposed approaches to address LEARNrsquo;s challenges including (1) the significance of temporal dimension (e.g., operating in different time regime), (2) new nodes entering or nodes leaving the network, (3) disruptions, anomalies and unpredicted behaviors and surprises, (4) diverse nature of nodes, their resources, and requirements, (5) periodic patterns (e.g., orbital info.), and (6) mission schedules. Our proposed solution is founded on our ongoing work to design and demonstrate a AI solution capable of outperforming traditional communication networking protocols. Finally, we propose to develop a MADRL (Ray RLlib)-CORE -based testbed to demonstrate our algorithms.