TY - GEN
T1 - CDDQN based efficient path planning for Aerial surveillance in high wind scenarios
AU - Dalai, Sagar
AU - O'Connell, Eoin
AU - Newe, Thomas
AU - Trslic, Petar
AU - Manduhu, Manduhu
AU - Irfan, Mahammad
AU - Riordan, James
AU - Dooly, Gerard
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Robotic surveillance, monitoring, and maintenance problem are open-for-research domains required by the military, industrial facilities, ports, airports, and various indoor and outdoor venues each having different needs. Recent work in path planning of aerial robotics is an emerging field of the surveillance problem, particularly for unstructured or unexplored areas. The nature of path planning problems with different foreign elements like wind, rain, and others escalate the cost of complex computation and power consumption. Due to constraints in payload and endurance, algorithms based on pose-graph, both from the run-time and solution point of view become inefficient when working with unstructured spaces. We propose a simple but effective Clipped Double Q-learning [1] based deep reinforcement learning algorithm (CDDQN)for efficient path planning under the influence of wind and with improved computational efficiency for surveillance in a port area. In the proposed algorithm we have formulated a dense reward structure in consideration of wind's effect on power consumption and time to reach the destination which led to a robust path planning system for high wind scenarios.
AB - Robotic surveillance, monitoring, and maintenance problem are open-for-research domains required by the military, industrial facilities, ports, airports, and various indoor and outdoor venues each having different needs. Recent work in path planning of aerial robotics is an emerging field of the surveillance problem, particularly for unstructured or unexplored areas. The nature of path planning problems with different foreign elements like wind, rain, and others escalate the cost of complex computation and power consumption. Due to constraints in payload and endurance, algorithms based on pose-graph, both from the run-time and solution point of view become inefficient when working with unstructured spaces. We propose a simple but effective Clipped Double Q-learning [1] based deep reinforcement learning algorithm (CDDQN)for efficient path planning under the influence of wind and with improved computational efficiency for surveillance in a port area. In the proposed algorithm we have formulated a dense reward structure in consideration of wind's effect on power consumption and time to reach the destination which led to a robust path planning system for high wind scenarios.
KW - Clipped Double Reinforcement Learning
KW - Deep Reinforcement Learning
KW - Path Planning
KW - Port Infrastructure Inspection
KW - Surveillance
KW - UAVs
UR - http://www.scopus.com/inward/record.url?scp=85173680150&partnerID=8YFLogxK
U2 - 10.1109/OCEANSLimerick52467.2023.10244322
DO - 10.1109/OCEANSLimerick52467.2023.10244322
M3 - Conference contribution
AN - SCOPUS:85173680150
T3 - OCEANS 2023 - Limerick, OCEANS Limerick 2023
BT - OCEANS 2023 - Limerick, OCEANS Limerick 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2023 OCEANS Limerick, OCEANS Limerick 2023
Y2 - 5 June 2023 through 8 June 2023
ER -