Intelligent Automation, Inc. — Department of Defense STTR Phase I: AF20C-TCSO1

Description

Currently, most state-of-the-art resource utilization algorithms are based on solving optimization problems. However, such approaches may require complete information of the system parameters, which may be difficult to obtain especially in highly dynamic airborne networks, and also may lead to high complexity and scalability concerns. High complexity can make it more difficult to obtain the optimal solution in real time, weakening the practicality of optimization techniques in airborne networks. This has led to interest in learning-based radio resource allocation strategies. We will develop robust, reliable and scalable distributed deep reinforcement learning (RL)-based power control and dynamic channel access policies for airborne networks (such as UAV networks). We will first develop RL-based radio resource allocation algorithms by considering both centralized and distributed training. Then, we will consider interference effects, develop jamming and adversarial attacks aimed at minimizing the accuracy of deep RL-based resource utilization, analyze the sensitivity of deep RL-based radio resource utilization and develop defense mechanisms. We will evaluate the algorithms under scenarios that are of interest to the AFRL using simulations and hardware-in-the-loop network channel emulation tests.