Stable Multi-Drone GNSS Tracking System for Marine Robots

TL;DR

This paper tackles the challenge of persistent marine robot localization when GNSS is unreliable near the surface by introducing a scalable offboard solution using multiple drones. The approach combines vision-based marine-robot detection (YOLOv11) and ByteTrack, GNSS triangulation from multi-view detections, and a confidence-weighted EKF with a hybrid IOU-GNSS data association, plus an inter-drone ID alignment mechanism. Key contributions include a cross-view ID consistency method, an estimation-aggregation EKF pipeline, and a translation ICP-based bias removal, demonstrated to achieve sub-2 m mean tracking error (0.94 m with three drones) and real-time performance on a Jetson Xavier. The results show that increasing aerial coverage improves accuracy, while the proposed methods maintain ID stability under wind-induced turbulence, enabling robust, low-cost, near-surface marine tracking for multiple robots and potentially broader marine monitoring tasks.

Abstract

Accurate localization is essential for marine robotics, yet Global Navigation Satellite System (GNSS) signals are unreliable or unavailable even at a very short distance below the water surface. Traditional alternatives, such as inertial navigation, Doppler Velocity Loggers (DVL), SLAM, and acoustic methods, suffer from error accumulation, high computational demands, or infrastructure dependence. In this work, we present a scalable multi-drone GNSS-based tracking system for surface and near-surface marine robots. Our approach combines efficient visual detection, lightweight multi-object tracking, GNSS-based triangulation, and a confidence-weighted Extended Kalman Filter (EKF) to provide stable GNSS estimation in real time. We further introduce a cross-drone tracking ID alignment algorithm that enforces global consistency across views, enabling robust multi-robot tracking with redundant aerial coverage. We validate our system in diversified complex settings to show the scalability and robustness of the proposed algorithm.

Figures (6)

• Figure 1: GNSS trajectory estimates of the same marine robot across two trials from Categories 1 and 2 in open-water bodies. The color gradient represents localization error, with brighter colors indicating larger errors. White arrows mark the direction of motion along each trajectory. The mean error is 0.842 m for Mission 1 and 1.282 m for Mission 2.
• Figure 2: Tracking system with three aerial drones collaboratively localizing two marine robots operating near the water surface.
• Figure 3: Sample frame from the drone at the highest altitude, showing two tracked marine robots (ID:1 and ID:2) along with two drones at lower altitudes in view.
• Figure 4: Geometric representation of the position estimation setup with three drones observing two marine robots. The image planes represent the cameras’ fields of view, while projection rays illustrate the triangulation geometry used for position estimation. For simplicity, Robot 2 is shown at the image center of all drones.
• Figure 5: Category 1 example showing ground truth and ICP-corrected trajectories, with black arrows indicating GNSS error corrections.