Design and Deployment of Resource-Aware Distributed Multi-Agent Algorithms

Distributed Unmanned Aerial Systems (UAS) are limited in computational resources, communication resources, and energy resources, which in turn drastically reduce their utility in multi-UAS applications. Orthodox countermeasures which include adding additional computational devices, advanced communication devices, or heavier batteries with more power, inversely correlate to the performance of the UAS. The reason being the added weight and the increased power requirements offset the additional resources the countermeasures provide. Hence, the feasible solution is to intelligently utilize the limited resources available. We present the resource-aware development of distributed multi-UAS control algorithms as the pathway toward intelligent resource utilization. This dissertation first introduces co-regulation techniques to dynamically allocate resources in distributed multi-agent systems controlled by consensus algorithms. Our need-based resource allocation shows significant savings in resources and a shorter time to convergence of the consensus algorithm whilst providing the user the option to adjust the controller gains for the user\u27s desired level of performance. We prove that our co-regulation techniques are robust to delays in communication. Our second contribution is a novel algorithm that combines consensus algorithms with active learning to drastically reduce the resource and time costs of re-training the convolutional neural network. Our final contribution is a series of resource-aware design decisions on the successful implementation of a hierarchical reinforcement learning-based linear quadratic integral (HRL-LQI) controller on a swarm of four UAS systems