Neural Networks Learning the Inverse Kinetics of an Octopus-Inspired Manipulator in Three-Dimensional Space

Abstract

The control of octopus-like robots with a biomimetic design is especially arduous. Here, a manipulator characterized by the distinctive features of an octopus arm is considered. In particular a soft and continuous structure with a conical shape actuated by three cables is adopted. Despite of the simple design the arm kinetics model is infinite dimensional, which makes exact analysis and solution difficult. In this case the inverse kinetics model (IK-M) cannot be implemented by using mathematical methods based on Jacobian matrix, because the differential equations of the direct kinetics model (DK-M) are non-linear. Different solutions can be evaluated to solve the IK problem. In this work, a neural network approach is employed to overcome the non-linearity problem of the DK-M. The results show that a desired tip position can be achieved with a degree of accuracy of 1.36% relative average error with respect to the total length of the arm.