Neural Network Based Controller for Visual Servoing of Robotic Hand Eye System

Abstract

Abstract — In this paper a neural network based controller for robot positioning and tracking using direct monocular visual feedback is proposed. The visual information is provided using a camera mounted on the end-effector of the manipulator. A PI kinematic controller is proposed to achieve motion control objective in the image plane. A Feedforward Neural Network (FFNN) is used to compensate for the robot dynamics. The FFNN computes the required torques to drive the robot manipulator to achieve desired tracking. The stability of combined PI kinematic and FFNN computed torque is proved by Lyapunov theory. Simulation results are carried out for 3 DOF articulated robot manipulator to evaluate the controller performance. Index Terms — Visual Servoing, Non-linear control systems, Feedforward Neural Networks, Perspective Projection.