Inverse kinematics evaluation for robotic manipulator using support vector regression and kohonen self organizing map

Abstract

Serial manipulators are designed as combination of serial links, which are connected using motor actuated joints. The absence of closed form solutions for manipulators leads to complex, time-consuming inverse kinematics analysis. In this paper, two different learning based accurate and relatively fast procedure for the calculation of all the joint angles for a specified given pose, is proposed for five DOF Dexter robotic manipulator. In first method, process of spatial decomposition is applied, and model parameters are estimated using a machine learning based method, support vector regression, that leads to less error and fast computing, and suitable for real-time applications. Second learning architecture is neural network based kohonen self organizing map (KSOM) method in which whole workspace discretion is done into three dimensional lattice and then clustered space is mapped using Taylor series expansion and gradient descent algorithm. Effective results using both learning based method have been shown for five DOF robotic arm.