Evaluating UAV Path Planning Algorithms for Realistic Maritime Search and Rescue Missions

TL;DR

The paper addresses UAV path planning for maritime search and rescue under uncertainty about target locations and environmental dynamics. It presents a Python-based framework that integrates OpenDrift-based target drift modeling with grid-based UAV planners and compares three strategies: Expanding Spiral, Boustrophedon, and a Global-Local Branch-and-Bound method, introducing a lightweight heuristic to reduce computation. A key contribution is the development of a practical, hybrid planner that fuses global region targeting with localized BnB search, tailored for real-time operation. Experimental results across varying distances show that spiral searches perform best at near-range, while the heavyweight but more robust BnB approach excels at longer ranges, offering a valuable guide for deploying UAV SAR systems in realistic sea conditions. The work provides a reusable, open-source framework and points to avenues for multi-UAV coordination and real-world deployment.

Abstract

Unmanned Aerial Vehicles (UAVs) are emerging as very important tools in search and rescue (SAR) missions at sea, enabling swift and efficient deployment for locating individuals or vessels in distress. The successful execution of these critical missions heavily relies on effective path planning algorithms that navigate UAVs through complex maritime environments while considering dynamic factors such as water currents and wind flow. Furthermore, they need to account for the uncertainty in search target locations. However, existing path planning methods often fail to address the inherent uncertainty associated with the precise location of search targets and the uncertainty of oceanic forces. In this paper, we develop a framework to develop and investigate trajectory planning algorithms for maritime SAR scenarios employing UAVs. We adopt it to compare multiple planning strategies, some of them used in practical applications by the United States Coast Guard. Furthermore, we propose a novel planner that aims at bridging the gap between computation heavy, precise algorithms and lightweight strategies applicable to real-world scenarios.

Figures (5)

• Figure 1: Example of a trajectory of the branch and bound agent right after it found a search target. Here, it succeeded in doing so after approximately 30 minutes after take-off. The search target's position is highlighted by a circle. The plot is taken from our framework.
• Figure 2: Two example plots taken from our framework. The agent performs the boustrophedon rectangle method. At the bottom, the agent is plotted with the trace of its trajectory for better overview of its performance. Recently visited grid cells are plotted in dark red while cells which were visited longer ago are brighter. At the top, the plot contains no trace to have a better look at the particles. The legends contain position, time, and found particles.
• Figure 3: Three output plots from the OpenDrift framework. It simulates four search targets, observable as four particle clouds in the left image, showing the start of the simulation. The middle and right images show the simulation after twelve and $24$ hours. The background shows the underlying water flow, changing over time. The location is roughly at $54.0 ~ \text{N}, 7.5 ~ \text{E}$.
• Figure 4: Schematic drawing of the three algorithms under investigation.
• Figure 5: An unfavorable case for boustrophedon search: The two images show two closely consecutive moments in a target search, as can be seen by the simulation time in the top right corner of either image. In the top image, the UAV (red and white square) is moving north in a straight line, then turning at the northern end of the search area to fly south afterwards -- this is shown in the bottom image. The search target (red particle, highlighted by a circle around it) moves west of the UAV's position while the drone is turning around at the northern edge of its rectangular search pattern. Plots are taken from our framework.