ABSTRACT
The present chapter argues for the general utility of coordinated control programs as a setting for the study of motor control in terms of programs involving the dynamic interweaving of activation of perceptual and motor schémas rather than a sterotypical list of movements. The notions of "virtual finger" and "opposition space" are introduced for the analysis of hand movements, with special emphasis on a theory of preshaping the hand in reaching for an object. The RS language for robot schémas is then outlined to make clear the notions of schema instance, instantiation, deinstantiation, and schema assemblage, which last includes the notion of coordinated control program. The rest of the chapter then addresses the issue of how schémas and schema assemblages may be implemented in neural networks. In some cases, a schema assemblage may be seen as no more than the anatomy linking the networks that implement the constituent schémas. In the case of learned behaviors, however, a more plastic implementation is required, and the notion of motor set is examined for clues as to how schema assemblages might be dynamically implemented in monkey and human cerebral cortex. Finally, an explicit "winner take all" neural network is presented for the task of positioning the wrist to achieve a desired opposition of a pair of virtual fingers; while back-propagation is used to train two simple artificial neural networks in choosing the appropriate (pad or palm) opposition and in allocating the appropriate set of real fingers to a virtual finger in grasping any one of a variety of objects.
