Direct homography control for highly maneuverable vehicles

Abstract

This paper introduces a new vision based controller that uses the entries of a homography matrix to control a highly maneuverable mobile robot. It aims to reduce the sensitivity of the control to calibration errors through connection of the perception and control in the data space of the sensor. The control law is formulated in a way that it is only dependent on features of the current image without measuring or calculating any position in the Cartesian space. Only the target image must be known and the target position should be estimated relatively to the feature plane. An evaluation of the robustness of the controller is presented and verified by simulation.