ABSTRACT
This paper reviews the theoretical picture that has emerged from a series of studies by which my colleagues and I have attempted to understand the interacting neural networks that may account for several dimensions of flexibility in primate sensory-motor control, especially voluntary arm control. These modeling studies have been guided by both psychophysical data and detailed information on biological neural networks. Many aspects of known anatomy (connectivity) and physiology (membrane and other transduction properties, excitatory or inhibitory sign of action) have now been incorporated into a comprehensive model. The model is built up from several distinct network modules, which correspond to forebrain, cerebellar, and spinal circuits. The behavioral range of the theory now includes key aspects of the following competencies: variable speed trajectory generation, size and speed scaleable handwriting production, independent control of joint angle and joint stiffness, automatic gain control of stretch (error) feedback, automatic force-pulse generation for velocity command tracking, self-organization of timing and gain of context-conditioned feedforward movement commands, self-organization of a 3-D egocentric coordinate system for representing external target locations, and self-organization of a direction-to-rotation mapping that allows efficient use of redundant degrees of freedom during visually-directed reaches.
