Artificial Intelligence: Computational Approach to Vision and Motor Control

Abstract

ConclusionsDURING THE PAST DECADE, it has become increasingly clear that to understand the brain we must study more than its biochemical and biophysical mechanisms and its outward perceptual and physical behavior. We also must study the brain at a theoretical level that investigates the computations that are necessaryko perform its functions. The control of movements such as reaching, grasping, and manipulating objects requires complex mechanisms that elaborate information from many sensors and control the forces generated by many muscles. The act of seeing, which intuitively seems so simple and effortless, requires information processing whose complexity we are just beginning to grasp. A computational approach to the study of vision and motor control has evolved within the field of artificial intelligence; this approach inquires directly into the nature of the information processing required to perform complex visual and motor tasks. This chapter presents a particular view of the computational approach and its relevance to experimental neuroscience. FOUNDING PRINCIPLES OF ARTIFICIAL INTELLIGENCEThe computational approach to vision and motor control is an outgrowth of the field of artificial intelligence, from which the basic tenets are derived (131). Research in artificial intelligence has two main goals: to develop computer systems that exhibit intelligent behavior and to understand the nature of intelligence itself. The field is founded on two basic principles: to separate the tasks performed by a complex information-processing system from the hardware that carries them out and to analyze natural intelligent systems through the synthesis of artificial systems that perform the same tasks.The birth of computers led to a distinction between a process specified by software and the machinery or hardware that executes the process. Computers have been built from a variety of components, including cams, relays, analog circuitry, transistors, and microchips, yet all are able to perform essentially the same computations. The thought naturally arises that neurons can be viewed as another form of computational machinery and that an intelligent process need not be limited to biological nervous systems but in principle could also be implemented in computers. If intelligent processes can be separated from hardware, then intelligence can be considered an abstract entity, subject to its own rules, laws, and structure, and can be studied in its own right. These laws are independent of the underlying computational machinery and reflect fundamental properties of particular information-processing tasks. 605

Constancy of central conduction delays during development in man: investigation of motor and somatosensory pathways.

“…From the age of 2 years the fastest conducting fibres in sensory and motor central pathways operate with constant conduction delays. If this phenomenon also holds for fibres of slower conduction velocity, it may serve to provide stability of timing for such processes as internal feedback and efference copy, which have been proposed to play major roles in motor control and learning (Berstein, 1967;Stelmach, 1976;Hildreth & Hollerbach, 1987;Nielson et al 1988). It is surprising that the staging of maturation is different for peripheral nerves, where early attainment of maximum fibre diameter leads to progressive changes in conduction delay with growth.…”

“…Sakamoto, Arissian & Asanuma, 1989) and in man (e.g. Bernstein, 1967;Stelmach, 1976;Hildreth & Hollerbach, 1987;Nielson, Neilson & O'Dwyer, 1988). Such integration involves not only peripheral reflex pathways, but also occurs at many levels within the central nervous system (Lundberg, 1982;Sakamoto et al 1989).…”

Optimization behavior of brainstem respiratory neurons

“…The human brain is capable of many complex computational tasks (Baron 1987;Hildreth and Hollerbach 1987;O'Keefe 1989). In motor control, remarkably optimal trajectories of simple movements can be accomplished through motor learning and adaptation, processes that are believed to be orchestrated by the cerebellum (Brooks 1986;Brooks and Thach 1981;Lisberger 1988).…”

Robotics and Neuroscience

“…Given the many layers of information processing in the primate nervous system, it is almost impossible to obtain a clear understanding of the underlying computations without a systems level view. It was primarily in the wake of David Marr's [71] seminal work on the computational neuroscience of vision in the 1970s and 80s that a computational approach to motor control was developed [72] -indeed, Marr himself proposed an influential model of motor learning for the primate cerebellum [73][74][75]. From a physics point of view, primates are inertia-dominated systems (i.e., Reynolds number >> 1), and such systems have been well studied in robotics, in particular in manipulator robotics, i.e., robots with arms and legs (as opposed to mobile robots on wheels).…”