Description

Abstract: This paper presents a soft robotic gripper for assisting persons with physical disabilities to feed them water. We have developed a thermoplastic polyurethane (TPU) gripper as well as a sevendegree-of-freedom robotic arm. Functions of object detections with YOLOv5 algorithm and user’s mouth detection based on an open-source library, MediaPipe, were implemented in the proposed system. Firstly, tasks of grasping and moving the detected object (a paper cup or a plastic bottle) to the mouth of the experimental user were conducted to evaluate the system. Secondly, tasks of grasping the plastic bottle placed horizontally on an experimental table and standing it up were carried out. Finally, we performed task of grasping the plastic bottle placed horizontally at a certain angle on the experimental table and standing it up. It was found from the experimental results that the object detections and the detection of the user’s mouth were successfully performed, that the success rates of these tasks were 80 % and more, and that the acceptable angles were obtained.

Keywords: soft robotic gripper; self-feeding system; people with physical disabilities; machine learning; YOLOv5; MediaPipe