Authors:

Lorenzo Demaria

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Recovering underwater vehicles such as the Submersible Autonomous Module (SAM) typically requires manual piloting of drones to position a hook for retrieval, especially in unpredictable ocean environments. This project supports a broader effort to automate that process entirely - from detecting the buoy to guiding the hook - using onboard vision and control systems. A convolutional neural network (CNN) has already been developed to recognize the buoy attached to SAM. However, autonomous capture also requires precise knowledge of the hook's position beneath the drone. This research addresses the challenge of estimating the 3D position of the swinging hook using either a monocular camera or a stereo camera setup, comparing the performance of both configurations alongside inertial measurements.. A Unity-based simulation replicates a drone equipped with a winch system, allowing dynamic rope control and realistic hook motion. Within this environment, a Robot Operating System 2 (ROS2) perception pipeline detects the hook in simulated images, reconstructs its 3D position using projective geometry, and fuses this data with inertial measurements through an Extended Kalman Filter (EKF). Our results show reliable hook tracking under slow flight conditions, with the EKF increasing robustness against visual dropouts. At higher speeds, performance is limited by camera frame rate and resolution. Future improvements include upgrading imaging hardware and optimizing detection for fast dynamics. This hook localization system is a critical step toward fully autonomous aerial recovery operations. If successful, it could reduce the need for manual piloting during Unmanned Underwater Vehicles (UUV) retrievals, increasing safety and efficiency in maritime missions. Keywords: Autonomous Aerial Recovery; Visual-Based Estimation; Dynamic Payload; Unity-Based Simulation; Extended Kalman Filter

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Lorenzo Demaria