NN-based solution of forward kinematics of 3DOF parallel spherical manipulator

In this paper, neural networks are trained to compute the forward kinematics of spherical parallel manipulator (PM) for laparoscopic surgery application. Instead of solving a set of nonlinear equations for the forward kinematics, neural networks are used to map the input angles of revolute joints to the orientation of the manipulator. The training data are obtained from inverse kinematic relationships and measured from the experimental prototype model of the manipulator. Levenberg-Marquardt algorithm is used to train the neural networks, which leads to the fast convergence of the networks. The trained neural network model of forward kinematics are used in the real time interface between the graphical model and a haptic device for the laparoscopes surgery training application. Simulation and experiments are carried out to verify the performance of the proposed method.