Home // International Journal On Advances in Software, volume 16, numbers 3 and 4, 2023 // View article

Deep Learning Decision Making for Autonomous Drone Landing in 3D Urban Environment

Authors:
Oren Gal
Yerach Doytsher

Keywords: Swarm; Visibility; 3D; Urban environment; autonomous landing.

Abstract:
Quadcopters are four rotor Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicle (UAV) with agile manoeuvring ability, small form factor and light weight – which makes it possible to carry on small platforms. Quadcopters are also used in urban environment for similar reasons – especially the ability to carry on small payloads, instead of using helicopters on larger vehicle which are not possible in these dense places. In this paper, we present a new approach for autonomous landing a quadcopter in 3D urban environment, where the first stage is based on free obstacle environment and maximal visibility for the drone in the palled landing spot. Our approach is based on computer-vision algorithms using markers identification as input for the decision by Stochastic Gradient Descent (SGD) classifier with Neural Network decision making module with greedy motion planner avoiding static and dynamic obstacles in the environment. We use OpenCV with its built-in ArUco module to analyse the camera images and recognize platform/markers, then we use Sci-Kit Learn implementation of SGD classifier to predict landing optimum angle and compare results to manually decide by simple calculations. Our research includes real-time experiments using Parrot Bebop2 quadcopter and the Parrot Sphinx Simulator.

Pages: 224 to 233

Copyright: Copyright (c) to authors, 2023. Used with permission.

Publication date: December 30, 2023

Published in: journal

ISSN: 1942-2628