Modelbased Visual Servoing Grasping of Objects Moving by Newtonian Dynamics

Abstract: Robot control systems are traditionally closed system. With the aid of vision, visual feedback is used to guide the robot manipulator to the target in a similar manner as humans do. This hand-to-target task is fairly easy if the target is static in Cartesian space. However, if the target is dynamics in motion, a model of this dynamical behaviour is required in order for the robot to predict or track the target trajectory and intercept the target successfully. One the necessary modeling is done, the framework becomes one of automatic control. >p In this master thesis, we present a model-based visual servoing of a six degree-of-freedom (DOF) industrial robot in the manner of computer simulation. The objective of this thesis is to manoeuvre the robot to grasp a ball moving by Newtonian dynamics in an unattended and less structured three-dimensional environment. >p Two digital cameras are used cooperatively to capture images of the ball for computer vision system to generate qualitative visual information. The accuracy of the visual information is essential to the robotic servoing control. The computer vision system detects the ball in image space, segments the ball from the background and computes the ball in image space as visual information. The visual information is used for 3D reconstruction of the ball in Cartesian space. The trajectory of the thrown ball is then modeled and predicted. Several ball grasp positions in Cartesian space are predicted as the thrown ball travelling towards the robot. At that same time, the inverse kinematics of the robot is also computed and it steers the robot to track the predicted ball grasp positions and grasp the ball when the error is small. In addition, the performance and robustness of this model-based prediction of the ball trajectory is verified with graphical analysis.