Planning Focus of Attentions for Visual Feedback Control

Planning functions for effective use of vision sensors are essential to make a hand-eye system work autonomously in the environment. This paper presents such a planning function to focus attentions so that accurate and reliable visual feedback tasks may be achieved. The system automatically generates a layout plan of windows called Visual Motion Cells which allow effective and flexible connections between visual information of the target object and the manipulator control. The layout planning consists of local and global selections of the object edges suitable to affix windows. The selections are performed based on the models of the object and the robot environment. The local selection of edges checks the minimum requirements such as contrast and length of the edges. Then the global selection determines the best combination out of the locally selected edges. A singular value analysis which evaluates expected accuracy in deriving control parameters based on the VMCs is utilized in the selection. Experimental results of the 6-DOF visual feedback control, using an actual hand-eye system, demonstrate not only effectiveness of the system but also importance of sensor planning in making intelligent robot systems with high autonomy in the future.