Investigating three learning algorithms of a neural networks during inverse kinematics of robots

摘要

The algorithms used in this paper was investigated while learning inverse Kinematic solution for a manipulator. In this paper, the performance of the three learning algorithms; Levenberg-Marquardt algorithm, Bayesian regularization algorithm, and scaled conjugate gradient algorithm is studied while learning single hidden layer network how to solve the inverse Kinematic problem of a three degree of freedom robot manipulator. During tests, Bayesian Regularization algorithm was found the best method among other methods with more accurate results but consuming larger time. Also, it is found that Scaled Conjugate Gradient method is inappropriate for the inverse Kinematic problem.