Visual Sensing Algorithm for Soft Manipulator based on Self-Organizing Map

Abstract

Due to the continuous deformation characteristics of the soft manipulator, the sensor used for the rigid manipulator is no longer suitable for sensing the shape, which is required for subsequent closed-loop control. In this paper, a vision-based shape sensing algorithm for soft manipulators is proposed, which does not need to mark points in advance or use the 3D point cloud of the target. The algorithm uses self-organizing map network to cluster the target obtained by the stereo camera, and then performs 3D reconstruction on the cluster centers to get a series of 3D points containing shape and position information. Experiments were carried out and the parameters and errors of the algorithm are analyzed in detail.