Framework for MR-iFLY Autonomous UAS Navigation and Collision Avoidance

Growing interest in unmanned aerial system (UAS) solutions for 6G communication networks stems from recent advancements in autonomous modules with machine learning computation and embedded graphical processing units GPUs. These technologies significantly improved UAS capabilities, however, providing a robust and operable autonomy module remains a significant challenge. The paper presents and evaluates a novel modular framework for UAS autonomy: MR-iFLY and discusses directions for its extension to enable 6G swarm solutions. We start by presenting the machine learning problem in implementing UAS autonomy on resource-limited devices and provide an in-depth overview of MR-iFLY's new depth estimation and collision avoidance technology. A comprehensive analysis and evaluation of MR-iFL Y's open-source modular framework including several performance indices are also presented. The results show that combining state-of-the-art machine vision components and GPU acceleration leads to a speedup of 15X over baseline models. We provide visualization of the MR-iFLY flight demonstration video obtaining vision-based collision avoidance. The obtained empirical results coupled with the current state of the industry suggest that MR-iFLY is a strong candidate to enable 6G communication swarms by providing autonomous collision avoidance and navigation capabilities while reducing the significant communication overhead.