Vision-Based Control of Robotic Arm Mounted on Unmanned Surface Vehicle for Rocket Recovery

Unmanned surface vehicles (USVs) equipped with robotic arms can perform various complex tasks such as grasping, sampling, and infrastructure repair. USVs are widely used for environmental monitoring, search and rescue missions, maintenance and repair, marine exploration, and its commercial applications. This paper presents an innovative vision-based strategy that utilizes a manipulator integrated with a USV to recover vertical takeoff vertical landing rockets in oceanic environments. This study focuses on a 4.3-m-long, 150-kg, three degree-of-freedom (DOF) WAM-V14 with a four-DOF manipulator. To detect and localize the landing rockets, a depth camera sensor and an advanced You Only Look Once Version Eight (YOLOv8) object-detection algorithm were employed. To train the YOLO algorithm, a customized bespoke dataset was developed for rocket identification. As the rocket descends and hovers, the system estimates its orientation and distance using a trained model. The measured distance served as the desired trajectory for the controller, thus guiding the USV toward the identified rocket and controlling the manipulator to reach its location. The integration of camera sensors, robotic arm manipulation, autonomous navigation, YOLO-based object detection, and a sliding-mode control system is promising for realizing affordable and safe rocket recovery operations in challenging oceanic environments.