The computational problem of motor control

Motor control systems are complex systems that process information. Orientation behaviour, posture control, and the manipulation of objects are examples of motor control systems which involve one or more sensory modality and various central neural processes, as well as effector systems and their immediate neuronal control mechanisms. Like all complex information processing systems, they must be analysed and understood at several different levels (see, e.g., Marr & Poggio 1977). At the lowest level there is the analysis of basic components and circuits, the neurons, their synapses, etc. At the other extreme, there is the study of the computations performed by the system -- the problems it solves and the ways that it solves them -- and the analysis of its logical organization in terms of its primary modules. Each of these levels of description, and those in-between, has its place in the eventual understanding of motor control by the nervous system. None is sufficient, nor is there any simple translation from one to another. A purely biophysical investigation, however exhaustive, can say nothing by itself about the information processing performed by the system, nor, on the other hand, can an understanding of the computational problem which the system solves lead directly to an understanding of the properties of the hardware. Two examples of motor control theories belonging to different levels will illustrate this point.