Experimental study on robot visual tracking using a neural controller

Visual tracking of a moving object has been an active research topic in the field of robotics and computer vision. In this paper, an experimental study is presented on a neural control design of a robotic manipulator to track a moving object using visual information. The proposed system integrates CCD visual data into an artificial neural network (ANN) for robot arm motion control. This design strategy features a fast and efficient control approach where the computation load is reduced to fit the real-time requirement. Integrated experiments have been carried out using a Mitsubishi RV-M2 industrial robot equipped with a CCD camera. After training the ANN controller with experimental data, the transformation from world coordinate to the robot coordinate can be eliminated. Robot motion control can be achieved without solving inverse kinematics of the manipulator. Furthermore, the proposed visual tracking system dose not require calibration data of the camera. The factors affecting tracking performance is analyzed and discussed.