Autonomous UAV Navigation in Wilderness Search-and-Rescue Operations Using Deep Reinforcement Learning

Muhammad Talha, Aya Hussein, Mohammed Hossny

Research output: A Conference proceeding or a Chapter in BookConference contributionpeer-review

1 Citation (Scopus)

Abstract

Wilderness Search and Rescue (WiSAR) operations require navigating large unknown environments and locating missing victims with high precision and in a timely manner. Several studies used deep reinforcement learning (DRL) to allow for the autonomous navigation of Unmanned Aerial Vehicles (UAVs) in unknown search and rescue environments. However, these studies focused on indoor environments and used fixed altitude navigation which is a significantly less complex setting than realistic WiSAR operations. This paper uses a DRL-powered approach for WiSAR in an unknown mountain landscape environment. To manage the complexity of the problem, the proposed approach breaks up the problem into five modules: Information Map, DRL-based Navigation, DRL-based Exploration Planner (waypoint generator), Obstacle Detection, and Human Detection. Curriculum learning has been used to enable the Navigation module to learn 3D navigation. The proposed approach was evaluated both under semi-autonomous operations where waypoints are externally provided by a human and under full autonomy. The results demonstrate the ability of the system to detect all humans when waypoints are generated randomly or by a human, whereas DRL-based waypoint generation led to a lower recall of 75%.

Original languageEnglish
Title of host publicationAI 2022
Subtitle of host publicationAdvances in Artificial Intelligence - 35th Australasian Joint Conference, AI 2022, Proceedings
EditorsHaris Aziz, Débora Corrêa, Tim French
PublisherSpringer
Pages733-746
Number of pages14
ISBN (Print)9783031226946
DOIs
Publication statusPublished - 2022
Externally publishedYes
Event35th Australasian Joint Conference on Artificial Intelligence, AI 2022 - Perth, Australia
Duration: 5 Dec 20229 Dec 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13728 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference35th Australasian Joint Conference on Artificial Intelligence, AI 2022
Country/TerritoryAustralia
CityPerth
Period5/12/229/12/22

Fingerprint

Dive into the research topics of 'Autonomous UAV Navigation in Wilderness Search-and-Rescue Operations Using Deep Reinforcement Learning'. Together they form a unique fingerprint.

Cite this