Online Learning Based Shape Control for a Soft Manipulator Based on Spatial Features Feedback

Abstract

Although soft manipulators are endowed with compliance and flexibility, most control strategies focus on end-effector control and lack shape control ability. This letter aims to design a shape controller for the soft manipulator. Firstly, we establish a modified forward kinematics model (FKM) based on the long-short-term-memory (LSTM) neural network to describe the mapping between actuation inputs and spatial features. The spatial features consist of the backbone curve and contour features. The backbone curve is represented by the piecewise Bézier curve under geometrically continuous constraint. The contour features are extracted from the camera-generated point cloud. Besides, an adaptive online learning based shape controller (OLSC) is designed by online back-propagating shape error. The stability of OLSC is proved based on the Lyapunov theorem. Finally, the random excitation model validation experiment demonstrates the prediction accuracy of the proposed modified FKM, and the shape control experiments in air and water validate the effectiveness of the proposed OLSC.