Motor Learning and Memory for Reaching and Pointing

To control reaching and pointing movements, the primate motor system draws on vision, audition, and other sensory modalities to estimate hand and target locations. We argue that the motor system represents these variables in visual coordinates, relative to the fixation point. According to a computational theory presented here, neural networks in the parietal cortex align information about muscle lengths and joint angles with an estimate of hand location relative to the fixation point. Related networks in parietal and frontal cortex, together with the cerebellum and basal ganglia, align the desired hand displacements, also in visual coordinates, with the joint rotations and forces needed to reach the target. The motor system updates these estimates when the eye changes orientation and whenever the hand or target changes location. Each network learns an internal model (IM) of the relationships among these sensorimotor variables and, in so doing, computes coordinate transforms and predictions about the limb's future state. In motor learning, IMs adapt to distortions in visual feedback and altered limb dynamics. This process begins with adaptation of existing IMs and results, after extensive practice, in the formation of new ones: the motor system has acquired a new skill.