Inverse Kinematic and Jacobian Solution for Serial Manipulator based on Optimized Neural Network

Singularities and uncertainty in manipulator dynamic is a major issue in kinematic control of manipulator which is obtained by applying robot model. In this paper, artificial neural networks with optimal training process and training data have been proposed as a way to solve this problem. The main idea of this approach is to use an artificial neural network to learn the characteristics of the robot instead of using an explicit model of the robot. Network is designed to have one hidden layer. The inputs are Cartesian position of the end-effctor along the X, Y, Z axis and Roll, Pitch and Yaw (RPY) orientation and linear velocity of end-effector. Network outputs is defined the position and angular velocity of each joint. In a workspace without obstacle, smooth geometric trajectorys in the joint space are designed. One of the major problems of using neural network is the selecting appropriate training data for network training, therefore, after presenting a method to generate training data, Levenberg-Marquardt Algorithm applied for training the network. At the end, results compared with other neural network and real values. The results show that the proposed neural network has good performance.