Open AccessCellular Network-Supported Machine Learning Techniques for Autonomous UAV Trajectory Planning
Author(s) -
Ghada Afifi,
Yasser Gadallah
Publication year - 2022
Publication title -
ieee access
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.587
H-Index - 127
ISSN - 2169-3536
DOI - 10.1109/access.2022.3229171
Subject(s) - aerospace , bioengineering , communication, networking and broadcast technologies , components, circuits, devices and systems , computing and processing , engineered materials, dielectrics and plasmas , engineering profession , fields, waves and electromagnetics , general topics for engineers , geoscience , nuclear engineering , photonics and electrooptics , power, energy and industry applications , robotics and control systems , signal processing and analysis , transportation
Autonomous trajectory planning is a hot topic in the UAV mission planning area of research. Autonomous UAVs have major use case applications which involve navigation in complex environments such as aerial photography, package delivery and relief operations. Many existing trajectory planning solutions rely on the GPS system. However, such GPS-based solutions do not provide a reliable real-time navigation solution, particularly in dense urban environments. Opportunely, cellular networks can be utilized as an attractive alternative for UAV navigation applications. We therefore propose to utilize existing 5G infrastructure to enable the UAV to navigate complex environments, independent of the GPS and other detectable signals transmitted by the UAV. Our objective is to propose an efficient solution to enable the UAVs to autonomously execute such tasks while meeting the real-time operational requirements, without the need to actually interact with the cellular network. For this purpose, we formulate the UAV trajectory planning problem as a joint objective optimization problem to minimize a composite cost metric that we introduce. The computational complexity involved in exact optimization techniques hinders obtaining the real-time calculation requirement that is needed due to the dynamic nature of the environment. To overcome this complexity, we utilize machine learning based techniques to solve the formulated trajectory planning problem. Specifically, we propose two machine learning-based techniques, namely, the reinforcement learning and the deep supervised learning-based approaches. We then analyze the performance of each of the proposed techniques as compared to the optimization-based approaches and other solutions from the literature. Our simulation results show that the proposed reinforcement and deep supervised learning-based solutions provide near optimal solutions to the formulated trajectory planning problem, with comparable accuracy of 99% and 98%, respectively, as compared to the optimal bound while meeting the real-time calculation requirement.