NEURAL REPRESENTATIONS FOR SENSORIMOTOR CONTROL .3. LEARNING A BODY-CENTERED REPRESENTATION OF A 3-DIMENSIONAL TARGET POSITION

A neural model is described of how the brain may autonomously learn a body-centered representation of a three-dimensional (3-D) target posit ion by combining information about retinal target position, eye positi on, and head position in real time. Such a body-centered spatial repre sentation enables accurate movement commands to the limbs to be genera ted despite changes in the spatial relationships between the eyes, hea d, body, and limbs through time. The model learns a vector representat ion-otherwise known as a parcellated distributed representation-of tar get vergence with respect to the two eyes, and of the horizontal and v ertical spherical angles of the target with respect to a cyclopean ego center. Such a vergence-spherical representation has been reported in the caudal midbrain and medulla of the frog, as well as in psychophysi cal movement studies in humans. A head-centered vergence-spherical rep resentation of foveated target position can be generated by two stages of opponent processing that combine corollary discharges of outflow m ovement signals to the two eyes. Sums and differences of opponent sign als define angular and vergence coordinates, respectively. The head-ce ntered representation interacts with a binocular visual representation of nonfoveated target position to learn a visuomotor representation o f both foveated and nonfoveated target position that, is capable of co mmanding yoked eye movements. This head-centered vector representation also interacts with representations of neck movement commands to lear n a body-centered estimate of target position that is capable of comma nding coordinated arm movements. Learning occurs during head movements made while gaze remains fixed on a foveated target. An initial estima te is stored and a VOR-mediated gating signal prevents the stored esti mate from being reset during a gaze-maintaining head movement. As the head moves, new estimates are compared with the stored estimate to com pute difference vectors which act as error signals that drive the lear ning process, as well as control the on-line merging of multimodal inf ormation.