UAV-Supported Maritime Search System: Experience from Valun Bay Field Trials

TL;DR

This study demonstrates an integrated autonomous maritime search system that tightly couples real-time drifter-informed surface flow reconstruction, adaptive probabilistic search planning, and YOLOv8-based object detection within UAVs. By fusing a dual OpenFOAM surrogate flow model, PSO-based boundary-condition fitting, and HEDAC-driven area coverage, the approach maintains robust search performance under environmental uncertainty. Field trials in Valun Bay show that adaptive diffusion guided by drifting error preserves high correspondence between modeled and actual target distributions, enabling reliable detections even with imperfect sensing and communication. The results highlight the practical viability of end-to-end autonomous maritime SAR systems and point to scalable strategies for future deployments and multi-UAV coordination.

Abstract

This paper presents the integration of flow field reconstruction, dynamic probabilistic modeling, search control, and machine vision detection in a system for autonomous maritime search operations. Field experiments conducted in Valun Bay (Cres Island, Croatia) involved real-time drifter data acquisition, surrogate flow model fitting based on computational fluid dynamics and numerical optimization, advanced multi-UAV search control and vision sensing, as well as deep learning-based object detection. The results demonstrate that a tightly coupled approach enables reliable detection of floating targets under realistic uncertainties and complex environmental conditions, providing concrete insights for future autonomous maritime search and rescue applications.

Figures (8)

• Figure 1: Illustration of the fused flow surrogate model concept combining bounded and open velocity fields. The optimization vector consists of $p$ and $\mathbf{w}_t$ values at boundary condition control points. The figure also displays measured and fitted velocity vectors at the drifter locations, used to establish the model fitting error.
• Figure 2: Example detections of targets, with associated detection confidence using YOLOv8
• Figure 3: Map of part of Europe with marked location of the experiment (red pin) and the location of the ICRA 2026 conference (green pin).
• Figure 4: Sea unit deploying the experimental target (left) and UAV base station overlooking the Valun Bay search domain (right)
• Figure 5: Example showing the floating target from the experiment (left) and the drifter used to obtain surface flow velocities (right)