INFANT neural controller for adaptive sensory-motor coordination

Abstract

This review presents a theory and prototype for a neural controller called INFANT that learns sensory-motor coordination from its own experience. Three adaptive abilities are discussed: locating stationary targets with movable sensors; grasping arbitrarily positioned and oriented targets in 3D space with multijoint arms, and positioning an unforeseen payload with accurate and stable movements despite unknown sensor feedback delay. INFANT adapts to unforeseen changes in the geometry of the physical motor system, the internal dynamics of the control circuits and to the location, orientation, shape, weight, and size of objects. It learns to accurately grasp an elongated object with almost no information about the geometry of the physical sensory-motor system. This neural controller relies on the self-consistency between sensory and motor signals to achieve unsupervised learning. It is designed to be generalized for coordinating any number of sensory inputs with limbs of any number of joints. The principle theme of the review is how various geometries of interacting topographic neural fields can satisfy the constraints of adaptive behavior in complete sensory-motor circuits.