We present an artificial neural network (ANN) that provides a unified approach to the problem of controlling saccades as well as smooth pursuit eye movements. Rather than trying to reproduce experimental results quantitatively, we focus on the functional requirements for such a system. We demonstrate that a fully connected, single-layer network (similar to the type suggested by Amari [Amari, 1997]) is capable of performing and controlling both kinds of movements under real-world conditions, given an appropriate preprocessing of visual input. A first-order approximation allows for some analytical treatment of smooth pursuit movements and reveals four different parameter regimes one of which is well suited to perform smooth pursuit movements even with unknown objects in dynamic real-world scenes.
展开▼
