Tobias Rodemann, Frank Joublin, Christian Goerick, "Continuous and robust saccade adaptation in a real-world environment", KI-Künstliche Intelligenz, vol. 03, pp. 23–26, 2006.

To saccade with a camera system to an interesting visual stimulus requires a mapping from image coordinates (pixel positions) to motor coordinates (e.g. pan and tilt angles). Because this relation can depend on many parameters, some of which might be varying over time, it is advisable to learn this mapping. However, most of the existing solutions require a special calibration procedure or at least a cooperating environment. In this work we describe a system, inspired from the human oculo-motor system, that can continuously adapt the saccade control in a real-world environment. The focus is not on speed of adaptation or the precision of saccades but on the robustness of the adaptation process. Our system can operate in feature-rich environments, monitor its own performance, and is not impaired by external motion.