seminars in THE NEUROSCIENCES, Vol 3, 1991 : pp 1 -2

Introduction : Sensori-motor integration

David Sparks

MANY NEUROPHYSIOLOGICAL studies are concernedwith how the neural events set in action by physicalstimuli are transformed into commands to modifyongoing movements or initiate new ones . As isevident from the papers in this issue, sensori-motorintegration is not a simple problem . Some of thecomplications considered in this issue are discussedbelow .

The control of even simple behaviors may dependon complex processing of sensory data . The jammingavoidance response of weakly electric fish requiresonly an increase or decrease in the frequency of thefish's electric discharge, yet the decision about whichto do is based on complicated computations of signalsfrom electroreceptors from different regions of thebody surface (Heiligenberg) .

Compensation for sensory perturbations may oftenbe achieved by more than one set of effectors . Legmovements compensate for changes in visualstimulation produced by tilting the surface on whicha fly is standing . Similar changes in the pattern ofvisual stimulation may occur during flight when headand body position are altered by wind currents . Butin this case, compensatory movements of the wingscorrect for the perturbation in the pattern of visualstimulation (Hengstenberg) . State-dependent gatingof sensory signals to different sets of effectors is apoorly understood process .

Conversely, the same behavioral response may becontrolled by sensory signals from several modalities .du Lac, Sparks, and van Gisbergen all consider theinteresting question of how visual signals coding thelocus of retinal stimulation, auditory signals codinginformation about interaural differences in the timingand intensity of sound waves, and somatosensorysignals coding the site of stimulation in a body-centered frame of reference can share motor circuitryfor the generation of orienting movements of thehead or eyes .

Despite these and other complexities, considerableadvances have been made in understanding the neuralmechanisms underlying sensorimotor integration .

Department of Psychology, University of Pennsylvania,3815 Walnut Street, Philadelphia, PA 19104-6196, USA©1991 by W. B. Saunders Company

1

Progress seems to be most rapid when the researchis guided by experimentally testable models (vanGisbergen) or when behavioral experiments carefullydelineate the exact neural computations to beperformed (Heiligenberg) . An analysis of sensori-motor integration usually begins with a descriptionof the quantitative relationship between the physicalproperties of sensory stimuli and the pattern ofneuronal activity observed at successive synapticrelays in the afferent pathways . In some modelsystems, the pattern of motor neuron activity thatprecedes specific movements and signals carried byneurons serving as inputs to motor neurons havebeen described in detail (van Gisbergen) . But onlyfor the jamming avoidance response in electric fish(Heiligenberg) is there a general understanding ofthe computations occurring at all the intermediatesteps involved in translating a receptor response intosignals carried by motor neurons .

Central to the issue of sensorimotor integrationis the problem of coordinate transforms . Complextransformations of sensory data are usually requiredbecause the spatial coordinates of afferent signalsare different from the spatial coordinates of themovements they guide. Kalaska provides a detaileddiscussion of this issue . He notes that even aftersensory signals have been translated into motorcoordinates, it may be difficult to determine theparameters of a movement to which a neuron'sactivity is related . Variables such as movementdirection, velocity and force are not independentbecause they are `inextricably linked by complexviscoelastic length/tension properties of muscles' .Kalaska also points out that a variety of signaltransforms may be required at different stages ofmotor control and he considers the possibility thatsignal transforms may be accomplished by graduallyaltering synaptic weightings across a series ofneurons, rather than in the discrete computationalsteps implied by hierarchical models .Many sensory signals and motor commands are

represented by the spatial and temporal patterns ofneural activity in maps of sensory and motor space .How the various maps are formed, what is representedin them, and how information is extracted from thepattern of neural activity in a map are all considered

2

in this issue . Some sensory maps are derived froma point-to-point mapping of the receptor surface, forexample, the electrosensory maps in the torus of theelectric fish (Heiligenberg), retinotopic maps in theoptic tectum (du Lac) and somatosensory maps inthe cortex of monkeys (Kalaska) . Other maps arebased on neural computations, such as the represent-ation of auditory space in the tectum of the barn owl(du Lac) and kinematic maps in parietal cortex ofmonkeys (Kalaska) . The maps facilitate additionalneural computations . For example, in weakly electricfish, the pattern of activation of electroreceptors onthe body surface is preserved in topographical mapsand timing signals from different regions of the mapare combined to compute signals of differential phasemodulations (Heiligenberg) .

Maps are also used for programming movements .Systematic variations in the direction and amplitudeof movements of the eye (Sparks, van Gisbergen) andhead (du Lac) are represented topographically in thetectum . Although an explicit map has not beenobserved, the direction of arm movements isrepresented in the motor cortex by the pattern ofactivity in a large population of neurons, each of whichfires maximally before movements in a particulardirection (Kalaska). The motor representations inthe tectum and motor cortex are higher order; thesignals observed here are quite different from thoseneeded by motor neurons sending signals to individualmuscles. How information for the formation ofdifferent types of motor commands is extracted fromthe spatial and temporal pattern of activity inmotor maps is discussed by Heiligenberg, Sparks,van Gisbergen and Kalaska . The problem ofcalibration and scaling of maps for appropriatematching of sensory and motor representations isconsidered by du Lac, who describes experimentsindicating that visual experience is critical for theprecise registration of sensory and motor maps inthe tecta of normal owls .

The concept of highly specialized commandneurons is noticeably missing in the articles presented

here. For the model systems explored, decisionmaking is not the task of a few specialized `commandneurons' but arises from the consensus in a largepopulation of neurons . I The direction and amplitudeof an eye movement is not determined by a few cellsthat code for a small, restricted range of movements .Instead, each member of the large population ofneurons active before a particular eye movementcontributes to the metrics of the movement (Sparks) .Similarly, the jamming avoidance response is drivenby a distributed system of contributions resultingfrom the evaluations of inputs from pairs of points ;there is no evidence for a central controller ordecision maker composed of a few command neurons(Heiligenberg) . And the direction of arm movementis thought to be based upon the average activity ofa large population of neurons in the motor cortex(Kalaska) .

The modern neurobiologist can approach theproblem of sensori-motor integration with a largearray of tools. A rich assortment of new methodsobtained from cell and molecular biology can be usedfor tracing anatomical pathways and studying thedevelopment of sensory and motor maps . Newcomputational tools and concepts are rapidly beingdeveloped and almost unlimited computational poweris available for the collection, reduction and analysisof data or for simulations and models of the neuralsystems being studied . Experimental questions canbe addressed using a variety of animal models andbehavioral paradigms . The challenge is to use theseresources in creative ways that will facilitate theunderstanding of both the computations requiredfor sensory-to-motor transforms and the cellularmechanisms used to implement these algorithms .

Reference

1 . Altman JS, Kien J (1987) A model for decision makingin the insect nervous system, in Nervous Systems inInvertebrates (Ali MA, ed), pp 621-643 . Plenum Press,New York

D. Sparks