An Underwater Robotic System with a Soft Continuum Manipulator for Autonomous Aquatic Grasping

Jiaqi Liu,Zhuheng Song,Yue Lu,Hui Yang,Xingyu Chen,Youning Duo,Bohan Chen,Shihan Kong,Zhuyin Shao,Zheyuan Gong,Shiqiang Wang,Xilun Ding,Junzhi Yu,Li Wen

DOI: https://doi.org/10.1109/tmech.2023.3321054

2024-01-01

Abstract:Delicate underwater manipulation tasks such as biological specimen collection are promising fields that require new robotic designs and intelligent robotic technologies. In this study, we proposed an automatic aquatic object-collecting system with a soft manipulator controlled by a reinforcement learning-based controller. For underwater sensing, we implemented a visual perception framework to restore the quality of the underwater image, detect the seafood animals, and track the target's position. The online learning ability of the reinforcement learning-based controller endowed strong adaptability for the soft manipulator against underwater disturbances. The water tank grasping tests show a 91.7% successful grasping rate without flow disturbance and 83.3% with flow disturbances. We demonstrated that the soft robotic collecting system gripped seafood animals in a lab aquarium as well as the natural seabed environment. The real-world experimental results showed that the robot successfully collected 28 shells within 40 min at a water depth of 15 m and even completed grasping tasks in a dark environment. Our results demonstrated that this manipulator prototype is potentially applicable for fully autonomous delicate objects underwater.