A neurobiologically motivated generalization of the adaptive model theory of human purposive movement to the control of nonlinear systems

Abstract

Adaptive Model Theory (AMT) is a computational theory and model of the information processing performed by the brain during voluntary movement. AMT proposes that the brain forms and adaptively maintains inverse dynamic models of the body and of the external environment. Previous AMT implementations have been restricted by their reliance upon linear inverse control techniques in the inverse modeling subsystems. We propose a nonlinear generalization of AMT capable of simulating human performance in the control of nonlinear external systems. A single-layer network of locally-recurrent dynamic neurons trained with feedback-error learning is used to form the inverse model. The structure offers a neurobiologically plausible means of extending the linear AMT architecture to account for human performance in the control of nonlinear systems.