In humans and many animals, images are stabilized on the retina to improve visual acuity. The vestibulo-ocular reflex (VOR) generates the eye movements required for stabilization during rapid head movements. The response generated by the VOR is a complex function of target distance and head rotation axis location. The goal of this dissertation was to understand the dynamical systems in the brain that allow the VOR to be accurately computed and adjusted during behavior.

Kinematic models of the reflex response were first derived mathematically and expressed in terms of eye position, vergence, head rotation and translation. The models describe the geometrical relationship between head motion and eye rotation and define the transformations performed on the sensory signals to construct the VOR responses. Dynamical models were constructed from the kinematic models and simulated on computer. These models closely reproduced the VOR velocity and acceleration responses to head velocity steps for different target distances and head rotation axis locations observed in primate (Snyder & King, 1992). The possible neural contributions of the vestibular nucleus, the cerebellum and other brain stem nuclei to the VOR are also proposed.

By defining a motor basis along the six extraocular muscles, the kinematic models were extended to derive a model with the precise VOR transformations from canal and otolith coordinates to motor coordinates. This model explains the observed responses of many VOR neurons combining canal, otolith and eye position signals. The model also makes predictions about other VOR neural responses.

A model of VOR gain adaptation was simulated on computer and verified an earlier cerebellar hypothesis by Miles & Lisberger (1981). The hypothesis suggested that long-term adaptation would occur in the brain stem vestibular pathway and not in the cerebellar cortex if the cerebellum had an inductive role in brain stem plasticity. To conform to physiological constraints, it was shown that the learning rate in the cerebellar cortex has to be greater than in the brain stem.

Snyder et al. (1992) showed that during vergence eye movements, the changes in VOR gain anticipate vergence changes. A predictive and adaptive cerebellar model was combined with our dynamical VOR model to reproduce these experimental findings. The cerebellar model received inputs from idealized vergence-disparity cells of the visual cortex and constructed a prediction of vergence angle to modulate the VOR gain predictively.