Vision-Based Tactile Perception for Soft Robotic Hand

Abstract

The innovative vision-based tactile perception for soft robotic hand is proposed in this paper. The soft finger consists of a coloured elastic inner chamber, an outer structure, an endoscope camera, and a sealing device. The camera is used to capture the images of the inner chamber for tactile perception of soft robotic hand. Firstly the design and fabrication of the soft hand is descripted. Then the algorithm of recognizing the bending posture is proposed. The color regions boundary of image is extracted, and the convolutional neural networks is deduced to estimate the bending posture. The referred points from the boundary are selected, and the bending direction and the object surfaces’ profile can be obtained by tracking displacement of the points. Finally, several experiments are designed and the results show the effectiveness of the proposed method and the superior performance of vision-based tactile perception.