TY - GEN
T1 - Autonomous UAV Navigation in Wilderness Search-and-Rescue Operations Using Deep Reinforcement Learning
AU - Talha, Muhammad
AU - Hussein, Aya
AU - Hossny, Mohammed
N1 - Funding Information:
Acknowledgement. This work is partially Council Grant DP200101211.
Funding Information:
This work is partially supported by the Australian Research Council Grant DP200101211.
Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - 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%.
AB - 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%.
KW - Autonomous navigation
KW - Curriculum learning
KW - Search and rescue
UR - http://www.scopus.com/inward/record.url?scp=85144820566&partnerID=8YFLogxK
UR - https://ajcai2022.org/program/
U2 - 10.1007/978-3-031-22695-3_51
DO - 10.1007/978-3-031-22695-3_51
M3 - Conference contribution
AN - SCOPUS:85144820566
SN - 9783031226946
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 733
EP - 746
BT - AI 2022
A2 - Aziz, Haris
A2 - Corrêa, Débora
A2 - French, Tim
PB - Springer
T2 - 35th Australasian Joint Conference on Artificial Intelligence, AI 2022
Y2 - 5 December 2022 through 9 December 2022
ER -