Delay/Disruption Tolerant Reinforcement Learning and Aurora Based Communication System

Stottler Henke proposes DREAMS, the Delay/disruption tolerant REinforcement learning and Aurora based coMmunication System, to address NASArsquo;s need for distributed autonomous adaptive network communications technology, including planning and scheduling. DREAMS-equipped devices will sense and receive data and utilize the latest advances in Reinforcement Learning to discover optimal link parameters (e.g., frequency, modulation scheme) to improve various metrics (focusing on maximizing throughput and minimizing BER). The best links are combined to form potential communications paths from a start node to an end node. When feasible, these paths are aggregated by a nearby scheduling node which consolidates the network state and then allocates communications through the network. The proposed work builds on a series of our relevant research efforts (including several NASA-funded efforts) involving resource planning and scheduling, machine learning, low-SWaP algorithms, and distributed algorithms. We have already built a graphical satellite communications simulator which uses NASArsquo;s WorldWind to display satellites and shifting communications links. In addition, we have already integrated our core planning and scheduling algorithm into NASArsquo;s core Flight System. For these reasons, we propose aggressive Phase I objectives culminating in a Phase I Prototype of TRL 4.