Enhancing Underwater Teleoperation with Simultaneous Estimation of Multi-DOF Hand Movements using Electromyography

Electromyography (EMG) of hand movements provides rich motion information and serves as a vital modality for teleoperation. This study introduces an EMG control paradigm based on simultaneous estimation of multi-degree-of-freedom (multi-DOF) hand movements, tailored for the teleoperation domain of Unmanned Underwater Vehicles (UUVs). We developed a multi-output model using machine learning to achieve estimation of multi-DOF hand movements. The model enables concurrent gesture classification and continuous prediction of wrist joint angles, facilitating simultaneous control of multi-DOF UUV motion through post-processing and behavior tree decision-making. Augmented by shared control functionalities, our framework embodies a harmonious fusion of human control and automated safeguarding protocols, ensuring the safe navigation of UUVs. Validation of the proposed method was conducted through an underwater teleoperation simulation experiment. Statistical analysis results demonstrate significant improvement in trajectory smoothness by nearly sevenfold compared with traditional gesture-based EMG control methods, along with reductions of 25.11% in task completion time and 5.89% in trajectory length.