Kinematic Calibration for Saccadic Eye Movements

The visual perception system of humanoid robots should provide sensorial information that fulfills the requirements imposed by perceptual tasks in natural environments. One prerequisite for such systems is the ability to observe the environment by actively moving its visual sensors. This ability allows to implement two essential behaviors for a cognitive visual system: smooth pursuit and saccadic eye movement. In the work presented in this paper we propose a kinematic calibration approach for the active camera system of the Karlsruhe Humanoid Head. The proposed method solves two fundamental problems when performing saccadic eye movements: the required kinematic model for open-loop control and the ability of stereo reconstruction with active cameras. We present experiments on the accuracy of the kinematic model, the stereo triangulation and the saccadic eye movement.