Configuration, Optimization and Management of DTN via Artificially Intelligent Decentralized Systems (COMMANDS)

We propose Configuration, Optimization and Management of DTN via Artificially Intelligent Decentralized Systems (COMMANDS). COMMANDS offers a decentralized AI-based solution to configure, optimize and manage communications for NASArsquo;s lunar networks. It will provide significant improvements in throughput, end-to-end latency, overhead and resource consumption in delay/disruption tolerant networks (DTN) backed by our previous NASArsquo;s SBIR program, LEARN. COMMANDS is based on our tactical AI design and builds a multi-agent deep reinforcement learning (MADRL) framework that enables the lunar network to sense, adapt, act and learn from its experiences and unknown environment autonomously. This automates and optimizes mission operation and resource efficiencies without necessitating involvement from a mission operations team. We propose to deploy graph-based machine learning that enables neural comprehension of all available network parameters and conditions into a node embedding. Next, our tactical AI uses a hybrid approach to combine expert systems and AI to bring robustness and intuition designs.COMMANDS extends our Python-based simulation of DTN Lunar communication for the training, testing, verification. We further use the CORE network emulation tool to demonstrate the validation and performance gains of the COMMANDS as a novel network traffic management software service.We propose an alternative approach for training graph neural network (GNN)-based design. To efficiently train GNN, first, we find the centralized solution to the network routing, resource allocation and control, bundle scheduling and fragmentation. The centralized solution will be formulated as mixed-integer linear programming (MILP). We propose to solve neural combinatorial optimization (NCO) to solve this centralized problem. The outcome is a centralized solution, which will help us in the design of a decentralized GNN-based solution.