Virtual Environment for Development of Visual Servoing Control Algorithms

摘要

Our problem considered was whether a virtual environment could be used for development of visual servoing control algorithms. We have used a virtual environment for the comparison of several kinds of controllers. The virtual environment is done in Java, and it consists of two industrial robots, IRB 2000 and IRB 6, a camera stereo system with two cameras mounted on the end-effector of the IRB 6, and one rolling ball and one bar. The experiment consists of tracking and grasping the ball using the different controllers. The robot IRB 2000 should grasp the rolling ball. The control of the robot is done in Matlab. We have three controllers. These controllers are function of the difference between the ball and the gripper. First, we use P-controller with a proportional gain. Second, the image-based Jacobian control is used but this controller needs an improvement because the robot tracks the ball with a little delay, then we use this controller with feedforward. The robot grasps the ball when the error between the ball and the gripper is less than one tolerance. In these two controllers, the depth is calculated with the two cameras (stereovision), therefore cameras need to be calibrated. Third, the hybrid controller is used. It is a mix of image-based and position-based controller. We use X and Y in Image space and Z in Cartesian space. Now, the 3D reconstruction is done from motion. It means we do not need calibrated cameras and the depth is calculated with adaptive control techniques. This adaptive control is used for recovering on-line the velocity of the ball. When the estimation of the ball is stable, the robot starts tracking the ball. (Less)