MAV Navigation in Unknown Dark Underground Mines Using Deep Learning

摘要: This article proposes a Deep Learning (DL) method to enable fully autonomous flights for low-cost Micro Aerial Vehicles (MAVs) in unknown dark underground mine tunnels. kind of environments pose multiple challenges including lack illumination, narrow passages, wind gusts and dust. The proposed does not require accurate estimation considers the flying platform as floating object. Convolutional Neural Network (CNN) supervised image classifier corrects heading MAV towards center tunnel by processing frames from single on-board camera, while navigates at constant altitude desired velocity references. Moreover, output CNN module can be used operator means collision prediction information. efficiency has been successfully experimentally evaluated field trials an Sweden, demonstrating capability different areas illumination levels.