A Robust Approach Toward Kernel-Based Visual Servoing

Abstract

We introduce a robust controller for kernel-based visual servoing systems. In such systems, visual features are sum of weighted-image intensities via smooth kernel functions. This information along with its derivative are input to the controller in which we have developed a sliding mode approach to generate system commands. In vision-based systems, image uncertainties affect the tracking performance and stability, and the target object may get out of the field of view. Unless considerable image uncertainties appear, such systems are able to track the object within desired precision. Hence, we have investigated the effect of the image noise as the main source of uncertainty, and encapsulated its characteristics in a proper representation. In order to fulfill the sliding condition, some bounds over image uncertainty and tracking errors are considered, and the controller gains are tuned online to keep the tracking error bounded. An application of the proposed method is experimentally tested on an industrial robot.