Visual servoing in robotic manufacturing systems for accurate positioning

Automated robotic manufacturing systems require accurate robot positioning. Visual servoing is an increasing popular method to enhance such positioning accuracy. Based on the error signal definition, visual servoing is classified into three approaches, Position Based Visual Servoing (PBVS), Image Based Visual Servoing (IBVS) and Hybrid Visual Servoing (HVS). In this research, firstly, a novel Neural Network (NN) based hand-eye calibration is introduced in PBVS. A MultiLayer Perceptron NN is used to approximate the nonlinear coordinate transform from image coordinates to real world coordinates in visual servoing. The main advantages of NN based hand-eye calibration are that it can solve the hand-eye calibration problem without estimating the hand-eye transformation and can improve the object tracking accuracy as well. The experimental results in an industrial manufacturing robot show that the proposed calibration method outperforms the current solving transformation matrix method and free hand-eye calibration method for 2D object tracking. Secondly, a new approach to switching control of IBVS with laser pointer is proposed. The simple off-the-shelf laser pointer is applied to realize the depth estimation. The proposed system is robust to the camera calibration and hand-eye calibration error, and is object model free as well. Comparing with traditional IBVS, it avoids image singularities and image local minima, and is successful for only partial image features in the field of view. Moreover, the trajectory of the robot end effector is shortened. The experimental results are given to verify the effectiveness of the proposed method in a robotic manufacturing system for assembl