Support Vector Regression based inverse kinematic modeling for a 7-DOF redundant robot arm

Abstract

In this paper, inverse differential kinematic modeling is performed for a 7-DOF (Degrees of Freedom) redundant robot arm. Two intelligent identification methods, namely Artificial Neural Networks (ANN) and Support Vector Regression (SVR) are used for modeling. The main strengths of SVR over ANN are that it doesn't get stuck at local minima and it has powerful generalization abilities with very few training data. An important problem in inverse kinematic solutions are the singularities which are points in the operational space where manipulator Jacobian is not invertible. In this paper, simulations are performed on a PA-10 model, to compare the modeling performances attained by ANN and SVR. It has been observed that SVR outperforms ANN in inverse differential kinematic modeling. Training data is obtained using direct differential kinematic equations of the manipulator and data points close to singularities have been discarded.