A Learning Framework to Inverse Kinematics of Redundant Manipulators

Abstract

Abstract This paper proposes a learning framework for solving the inverse kinematics (IK) problem of high DOF redundant manipulators. The latter possess more DOFs than those required to obtain the end effector (EE) pose. Therefore, for a given EE pose, several joint angle vectors can be associated. However, for a given EE pose, if a set of joint angles is parameterized, the IK problem of redundant manipulators can be reduced to that of non-redundant ones, such that the closed-form analytical methods developed for non-redundant manipulators can be applied to obtain the IK solution. In this paper, some redundant manipulator’s joints are parameterized through workspace clustering and configuration space clustering of the redundant manipulator. The growing neural gas network (GNG) is used for workspace clustering while a neighborhood function (NF) is introduced in configuration space clustering. The results obtained by performing a series of simulations on a 7 DOFs redundant manipulator demonstrate the effectiveness of the proposed approach.