!

!

Citation

MacIver, M. A. (2009). Neuroethology: From Morphological

Computation to Planning. The Cambridge Handbook of Situated

Cognition. P. Robbins and M. Aydede. New York, NY, Cambridge

University Press: 480-504.

CHAPTER 26

Neuroethology

From Morphological Computation to Planning

Makolm A. MacIver

Introduction

Neuroethology is a field devoted to under-standing the nervous system through thebroader contexts of evolution, natural his-tory, ecology, and everyday behavior; inother words, it is the study of situated ner-vous systems. Its focus is on how neural sys-tems subserve behaviors that an animal per-forms in its natural habitat, such as capturingprey and evading predators, finding a mate,and navigating through its domain. Oftenthe nervous system is examined in animalsthat exhibit extraordinary specializations inbehavior, such as sonar-emitting bats andfish that hunt by detecting changes in aweak, self-generated electric field, becausesuch specializations result in specializedneural circuitry, making neuron-level anal-yses more tractable. These experimentallytractable animals are sometimes referred toas "model systems." People generally preferto work with an established model systemso that they can build on a body of knowl-edge that has already been gathered aboutthe system.

As a combination of the laboratory sci-ence of neurobiology and the field observa-tion science of ethology, neuroethology hassignificant challenges, including determininghow much of the ecological context of ananimal is necessary or practical to importinto the laboratory (for a review, see Pfluger& Menzel, 1999). Once the animal and itsreduced environment are in the laboratory,another problem to be solved is how toextract, from its continuous and highly irreg-ular activity, a particular behavior to focuson. One way that this issue has been dealtwith historically (Pfluger & Menzel, 1999)is through avoidance of learned behaviors infavor ofhighly stereotyped innate behaviors.More recently, neuroethology has embracedmodel systems for studying limited forms oflearning, such as the zebra finch (Marler,1991), which learns its song from a tutoronly during the first thirty-five days of life.After this period, the song is "crystallized"and does not vary.With a behavior identified and character-

ized, the next step is to uncover its puta-tive neural basis. Here a new set of issues

NEUROETHOLOGY

comes into play, centered on the technicaldifficulty of obtaining reliable recordings ofthe electrical activity of the animal's ner-vous system. Often this may not be fea-sible unless the animal is under anesthe-sia. Options include working on a "reducedprep/' in which slices of brain, dissociatedcells, or whole parts of the nervous systemmay be placed into a chamber where theycan be perfused with oxygen, nutrients, orneuroactive drugs and stabilized for record-ing. A better approach, comparatively rarebecause of its technical difficulty, is implant-ing one or more permanent recording elec-trodes in the brain, mounted on a platformthat is glued to the head with the signals sentout via cables or wirelessly; this techniqueis called "chronic recording." Such tools atbest monitor the ongoing activity of one tohundreds of cells in networks that can con-sist of millions of cells (and only the electri-cal activity, at that). Larger-scale properties,spanning multiple networks, biomechan-ics, and behavior, are sometimes examinedthrough the development ofcomputer mod-els of individual components that are thenrecombined in computo in simulation envi-ronments, an endeavor sometimes referredto as "computational neuroethology" (Chiel& Beer, 1997; Cliff, 1995). This then is theneuroethologist's modus operandi: find ananimal offering certain experimental conve-niences such as behavioral specializations,identify and quantify a behavior, and mea-sure its neural correlates, if possible in the"awake, behaving animal," or if not, in somereduced preparation while delivering stimulisimilar to those occurring during the identi-fied behavior.Neuroethology has several aspects that

will be of interest to the situated cognitioncommunity. Because ethology is all aboutwhat goes on when an animal is embeddedin its environment (consider the title of animportant monograph ofone ofits founders:The Animal in Its World [Tinbergen, 1972]),and neurobiology is all about what goes on inthe brain, often with only cursory attentionto matters beyond the periphery ofthe head,how can the enterprise of neuroethology

work in the first instance? Even if you donot believe that the properties of interest inneuroethology relate to cognition, how thetension between the outside-the-body ten-dencies of ethology and the inside-the-headtendencies of neurophysiology plays outin neuroethology is alone instructive. Thatthese aspects can be in tension needs furtherexplanation; otherwise one might think thatthis is simply a matter of groups of scientistsworking on different parts of a big problem,where everyone recognizes that a division oflabor is practical and necessary. Neurophys-iological approaches often implicitly suggestthat many, perhaps most, components ofthe nervous system can be understood with-out examining an animal's larger context,be it behaVioral, biomechanical, or evolu-tionary. In contrast, neuroethologists believethat unless the larger context is understood- for example, by quantifying the profileof sen;ory signals that an animal is subjectto in its habitat or by placing neural char-acters of related animals into phylogenetictrees ("neurocladistics") - many aspects ofneural function will not be understood.Exemplifying this point, in several modelsystems neuroethologists have found thatneural circuits involved in sensory process-ing exhibit very different response proper-ties when they are subjected to naturalisticstimuli than to the nonnaturalistic stim-uli more commonly used in neurobiology,because these are easier to generate andmanipulate experimentally (see Sharpeeet al., 2006, and references therein).In view ofthese considerations, in the first

part of the chapter I will discuss how thetension between the inward- and outward-looking approaches of neuroethology maybe resolved through excising bits of theworld and encapsulating them into virtualreality apparatuses in the laboratory. In thesecond part of the chapter I will argue thatresults from neuroethology do, in fact, relateto cognition. Here I will first detail resultsfrom studies of morphological computationthat expose the computational role of shapeand structure in animal bodies in adaptivebehavior, and then I will describe some

MALCOLM A. MACIVER

recent results concerning the neuroethol-ogy of prey-capture behavior that may giveinsight into the origin of the paradigmat-ically cognitive faculty of planning. I willargue that the evolution of sensing systemsthat enable animals to perceive their envi-ronment far beyond the bounds of wherethey are immediately moving, possessing abuena vista, ifyou will, is key to the origin ofplanning. I call this the Buena Vista SensingClub hypothesis. Through planning, mem-bers of the Buena Vista Sensing Club areable to make better use of this space in theguidance of their behavior than their morereactive mala vista brethren.

1. Internalism, Externalism, andVirtual Worlds in Neuroethology

The methodological and philosophical basisof neuroethology takes the situatedness oforganisms as a basic fact and works forwardfrom there. Both traditional cognitive sci-ence and much of neurobiology differ inemphasis, focusing on what goes on in thehead (Clark, 1997; Hutchins, 1995; Noe, 2004;Rowlands, 1999; Wilson, 1995). This cran-iocentric approach often goes by the labelof "internalism." What is variously called"situated cognition," "embodied cognition"(Ballard, Hayhoe, Pook, & Rao, 1997; Hauge-land, 1998), "distributed cognition," "senso-rimotor accounts of perception" (O'Regan& Noe, 2001), "active externalism" (Clark& Chalmers, 1998), "enactive externalism"(Noe, 20°4), and "wide computationalism"(Wilson, 1994) all share the view that theproperties of interest can depend on thehead plus body plus environment. I will referto these different approaches as "external-ism."Although situatedness is basic to neu-

roethology, as alluded to in the introduc-tion, it is true that its two contributing dis-ciplines pull in different directions: ethologyto externalism and neurobiology to inter-nalism. How is this tension resolved inneuroethology? Consider investigations intothe sensory systems of two popular modelsystems in neuroethology, both nocturnal

animals that - like miners wearing head-lamps - provide their own source of illu-mination for perceiving their dark envi-ronments: sonar-emitting bats and electricfield-emitting fish (Figure 26.la and 26.1C).For both model systems a substantial partof the sensory system exists in the interplaybetween generated signals and objects inthe environment (Nelson & MacIver, 2006).Thus, a fundamental part of the literature isdevoted to describing and analYZing how theself-generated signals propagate and interactwith the environment - characterizing thesensory system as it operates outside of thebody of these animals (on bats, see Boonman& Jones, 2002; Ghose & Moss, 2003; Hartley&Suthers, 1989; Miller &Surlykke, 2001; Par-sons, Thorpe, & Dawson, 1997; Schnitzler,Moss, & Denzinger, 2003; on fish, see Assad,Rasnow, & Stoddard, 1999; Chen, House,Krahe, & Nelson, 2005; Rasnow, 1996; Ras-now & Bower, 1996). Recently, this work hasbeen referred to as research into the sensoryecology of an animal (Barth & Schmid, 2001;Dusenbery, 1992). In these studies, the ner-vous system of the animal may hardly bementioned.Whereas the externalist tendencies of

neuroethology have their most literal formin the context of this work on the active-sensing model systems of bats and weaklyelectric fish, the focus on the relevantaspects of the external world in behavior ispervasive in the discipline, from characteri-zation of odor plumes for studies of moths(Murlis, Willis, & Carde, 2000) to recon-struction ofwhat a fly sees while it is buzzingaround in an open field (van Hateren, Kern,Schwerdtfeger, & Egelhaaf, 2005)' As point-ed out by James J. Gibson (1979, p. 57), thepioneer of ecological approaches to percep-tion, mechanical signals in the environment(stress, strain, pressure, inertia, gravity, fric-tion, drag) are no less important in under-standing behavior than are sensory signalsin the environment. Neuroethologists there-fore have an increasing stake in investigat-ing mechanical factors with respect to boththe animal and the environment (e.g., Chiel& Beer, 1997; Dickinson, 1996; Full & Tu,1991; MacIver, Fontaine, & Burdick, 2004)'

10m

O.05m

100m

--

case. My argument is not that the importantproperties of interest depend on the brainalone, which intemalism would insist on (atleast for cognitive properties). Instead, oftenthe properties very definitely do dependon the brain plus body plus environment,but the elements of the causal chain thatare most resistant to being understood arelargely in the brain and body alone, in partfor mundane technical reasons such as acces-sibility to measurement. Evidence for thisclaim is that, in some cases, we have thecapability to mimic the animal's world insuch a way that the animal does not knowany better. I will present three examplesof virtual worlds used in neuroethologi-cal research that exemplify this point (Fig-ure 26.2) for studies of insects, fish, andmammals. These virtual worlds pass theembodiment Turing test - animals situatedin them happily converse with the proxy asif it were the real deal.The first example is from work at the

Max Planck Center for Biological Cybernet-ics, in Tiibingen, Germany, in the late 1960s.

b

d

O.05m

c

NEUROETHOLOGY

Figure 26.1. Sensory volumes of different active sensing animals.a. Bat echolocation beam; the illustrated range is the estimateddetection range for small prey (mosquitoes) averaged across severalbat species. b. Dolphin; the illustrated range is the estimateddetection range for a prey-sized, water-filled sphere. c. Weaklyelectric fish; the illustrated range is the estimated detection rangefor small prey (water fleas). d. Rat whisker system. Reproducedwith permission from Nelson and MacIver (2006).

In parallel with sensory ecology, I refer tothis as an animal's "mechanical ecology."Those with internalist leanings should not

fear, however. First, there is the safe refugeof the cognitive; to take this refuge, theintemalist would argue that what we areconsidering here are not cognitive proper-ties in the first instance, whereas only thenotion that cognition depends on the brainwas at issue. I will return to this point later.Aside from this defense from the irrelevanceof neuroethology, recall that the modusoperandi of the neuroethologist is to even-tually relate natural behaviors to neuronalstructures. This is done by recording neu-ral activity either in awake, behaving ani-mals or, more often, in fixed and anes-thetized, or otherwise drastically reduced,forms (e.g., brain slices). our internal-ist interlocutor will exclaim, "Now we get tothe important business: what you are reallyinterested in is what goes on in the brain;you're not stuffing recording electrodes intothe environment, are you?" There is somesense to the intemalist's intuition in this

MALCOLM A. MACIVER

../;,'

K > ....

-::. \'. ::-...... \...... '.

\.ar .. '.

so""

air

c

record

stimulate

b

Figure 26.2. Three virtual worlds in neuroethology. a. Reichardt'sclosed-loop apparatus for the study of optomotor responses inhouse flies (Reichardt & Wenking, 1969)' The fly is tethered on arigid rod that senses the torques generated by the fly as it steers leftor right and moves the surrounding visual panoramacomplementary to the motions that would naturally result fromthose torques. Modified from Reichardt and Poggio (1976). b. Aclosed-loop apparatus for the study of electromotor responses inweakly electric fish. The fish is held in the center of the tank eitherby mechanical constraint (not shown) or paralyzed by a drug andplaced into a holder (not shown). Two graphite rods parallel to thefish deliver an electric field similar to what a nearby fish wouldemit. By bringing the frequency of the delivered field close to thatof the fish in the tank (monitored via the vertical graphite rods), anelectromotor behavior, the jamming-avoidance response, istriggered and sustained by following the fish's frequency changes.Similar to the apparatus in Watanabe and Takeda (1963)' c. Aclosed-loop apparatus for studying the behavior of rats, based onDahmen's (1980) similar device for insects and GCitz and Gambke's(1968) servosphere. The ball is suspended by an air cushion. Theanimal walks while tethered to a force and torque sensor, whosesignals are measured to alter the real time video projected onto therat's visual surround. From Holscher et ai. (2005) with permission.

During that time, Werner Reichardt used apreexisting system for recording the turn-ing torque generated by flies that are fly-ing while rigidly affixed to the end of a thinrod (Figure 26.2a). In his apparatus, the mea-sured turning torque was transformed into arotation of the surrounding visual panoramawith an angular velocity proportional to thetorque (Figure 26.2a; Reichardt & Poggio,1976; Reichardt & Wenking, 1969), so thatwhen the fly turned to the left or right, thesurrounding visual scene changed appropri-ately. This approach has been heavily usedin studies of fly behavior, from the neu-ral basis of optomotor behaviors to learningand flight mechanics, as it enables precise

control of the stimulus and quantification ofbehavior.The second example involves the elicita-

tion of an interesting electromotor behaviorin weakly electric fish. These fish dischargetheir weak electric field at a particular fre-quency that is both species specific and indi-vidual specific. When two fish happen acrosseach other in their native habitat while dis-charging within a few hertz of each other,they will shift their discharge frequencies(via a motor nucleus in the brain) to avoidjamming their electrolocation systems. Tostudy this behavior, a fish is placed in atank and fixed in place either mechanicallyor by paralysis, using a drug (Figure 26.2b).

NEUROETHOLOGY

The fish's electric field is recorded, anda field that is slightly higher or lower infrequency than its field is introduced intothe tank, eliciting the jamming-avoidanceresponse (Watanabe & Takeda, 1963)' Theapproach used here is again closed loop,because any changes in output frequencyby the fish are monitored and used to ad-just the frequency of the jamming stimu-lus so that the behavior can be continuouslyelicited. Using this method combined withneurophysiology, the jamming avoidanceresponse became the first vertebrate behav-ior whose complete neural circuit from sen-sory input to motor output was understood(Heiligenberg, 1991).The third example, also originally devel-

oped in the late 1960s at the Max PlanckCenter, is from studies by Karl Gotz andothers: an insect would walk on a spherethat counterrotates in such a way that theanimal is always in one place, again greatlysimplifying the measurement of behaviorwith video and the delivery of controlledstimuli. The original version, called a "ser-vosphere," tracked the animal's location andfed the movement back to two motors thatrotated the sphere so that the animal stayedroughly at the apex of the sphere (Gotz &Gambke, 1968; Kramer, 1975; Varju, 1975)' Amore recent variation (Figure 26.2C) does notuse motors but instead has the animal teth-ered above the apex of a hollow sphere thatis suspended by an air cushion and rotatedby the animal's own movements (Dahmen,1980; Holscher, Schnee, Dahmen, Setia, &Mallot, 2005)' It comes as no surprise thattwo of these three examples stem from aninstitute devoted to cybernetics. For moredetails on this tradition see Eliasmith (thisvolume).One response to these examples would

be that this proves the internalist's point -the world does not matter; all one needs isto synthesize and input the right kind ofsignals. However, Alva Noe (2004, p. 224)makes a convincing case that far from being atriumph ofinternalism, the success ofvirtualreality scenarios is precisely what an exter-nalist should hope for: it demonstrates thatthe elicited behavior depends on the brain

plus the given part of the world. Neverthe-less, the externalist may be given pause byconsidering the success ofvirtual-reality sys-tems in neuroethology. These demonstratethat we understand the relevant couplingsto the external world in select cases in a suf-ficient - not to say complete - way, whereaswe have a long way to go to understand theinternal goings-on of the brain and body.

2. Is Neuroethology Relevantto Situated Cognition?

I had mentioned that one way a cogni-tively inclined person might assert thatneuroethology is not relevant to cognitionis by arguing that the processes under con-sideration are not cognitive. That is, giventhat many cognitive scientists hold that theextension of cognition itself outside of theusual human case is controversial (Wilson& Clark, this volume), how can results onthe zoo of small animals that neuroethol-ogy studies provide insight into situatedcognition? The difficulty of bridging thesetwo domains is exemplified by the mis-match between the cognitive capacities weattribute to humans with those attributesof nonhuman animals that are sanctionedby neuroethology. The attributions of neu-roethologists have a distinctly mechanisticflavor, such as "encodes stimulus amplitudewith neuron spike rate" or "rotates headto zero azimuth to target by comparingthe intensity of sounds between the ears"or, on a behavioral level, "advertises fit-ness to a potential mate by the complex-ity of the song." Cognitive faculties, suchas remembering, planning, and deliberating,have their clearest form in conscious, occur-rent thought processes in humans, and theirpresence in nonhuman animals is not oftenbroached by neuroethologists.Nonetheless, an attraction of external-

ist approaches such as situated cognitionis their ecumenical approach to all kindsof phenomena that have been previouslybarred from the tent of cognition. Accord-ing to Adams and Aizawa (this volume), thismore open approach is confused and results

L.. _

MALCOLM A. MACNER

from externalists (1) making the error ofslid-ing from "x is causally coupled to a cognitiveact" to "cognition is constituted byx," and (2)failing to clearly differentiate cognitive fromnoncognitive processes. An example theygive of the first error is to make the jumpfrom the fact that the reading of a ther-mostat is causally coupled to room temper-ature to the notion that the thermostat isconstituted by the room-thermostat system.Similarly, they would argue, the fact thatthe cognitive act of long division with paperis causally coupled to the piece of paperis a fallacious basis for asserting that cog-nition is constituted by the human-papersystem. However, unless one leaves cogni-tion as an unanalyzed whole, it will havecausally coupled subcomponents. Of theseparts it would be correct, presumably, tomake the move from causal coupling to con-stitution; in a case like this, the thermostatreally would be constituted by the room-thermostat system. The key issue, therefore,is the proper characterization ofwhat countsas cognition. Because this is poorly definedat present, I will take the strategy of consid-ering causal proximity to cognition, in thehope that although we may not have a cleardefinition of cognition, we have some intu-ition about when we are closer to it or fur-ther away from it. We have clear intuition,for example, that the central processing unitin a computer is more proximal to a com-putation performed by the computer than isthe table on which the computer is resting.Similarly, the most compelling examples ofsituated cognition are examples in whichsome bit of the world plays a central role:for example, the use ofexternal objects suchas maps during navigation (Hutchins, 1995),racks of letter tiles in Scrabble to rememberwords (Maglio, Matlock, Raphaely, Cher-nicky, & Kirsh, 1999), and paper and pen tomanipulate numbers (Zhang &Wang, 2005)'In these cases, one can reasonably argue thatdrawing the boundary ofthe cognitive at theskin is arbitrary (Clark & Chalmers, 1998).With these considerations in mind (ifnot,

perhaps, in the head), I will present resultsfrom neuroethology with a range of depen-

dencies on the external world and causalproximities to the cognitive. I should stateat the outset that this is not meant to bea representative set of results; most of neu-roethology consists of detailing the opera-tion ofneuronal circuits that support naturalbehaviors, and these will not be discussed.Even the preceding virtual-world examplesare not representative of neuroethology as awhole because the discipline has a strongbias toward sensory systems and sensoryacquisition, with only a few active systemsfor the study of motor function apart fromthe ones presented, such as research into theneural network responsible for chewing incrustaceans. This bias may be in part dueto there being too few closed-loop appa-ratuses like those described previously, allof which require a considerable amount ofengineering expertise. For a more represen-tative sampling of neuroethology, I refer thereader to Web sites ofneuroethology courses(Hopkins, 2005; MacIver & Nelson, 1996),surveys (Barth & Schmid, 2001; Carew, 2000;Hughes, 1999; Land &Nilsson, 2002; Zupanc,2004), and conference proceedings of theInternational Society of Neuroethology.Under the rubric of morphological com-

putation, I will give the example of howthe ears of bats perform a signal processingfunction by dint of their convoluted shape,and how the geometry of the fly compoundeye may allow for the efficient extraction ofself-motion from the optic-flow field. Thesetwo examples concern how the earlier men-tioned sensory ecology of an animal, theambient set of behaviorally relevant signalsin an animal's habitat, is interwoven withpreneuronal signal processing. I then brieflydiscuss bilateral symmetry, a ubiquitous fea-ture of animal body plans, and the passivewalker, a robotic model of human walkingthat consists of only rigid links and joints andyet is able to walk with a human gait. Bothare examples ofhow the mechanical ecologyof an animal (the ambient inertia, stiffnessand compliance, drag or contact friction, andso on, of the animal and its coupling to thehabitat) is interwoven with locomotion andcontrol of position.

NEUROETHOLOGY

Figure 26+ A potpourri of bat pinnae. Ear-shape data andrendering courtesy of Rolf Mueller. The pinna is the outer ear; thetragus is the small pointed flap of tissue coming up from the base ofthe pinna.

These two examples of morphologicalcomputation highlight the importance oftransneuronal processes in sophisticated sig-nal processing and mechanical capabilities.However, the capabilities at issue are distantfrom more familiar examples of cognition.Following the section on morphologicalcomputation, I try to partially close the gapby exploring how recent work on the neu-roethology of prey capture may give insightinto one potential origin of the paradigmat-ically cognitive faculty of planning. On thebasis of this work, I will argue that the sen-sory transcendence of the space of imme-diate movement provides a core basis ofplanning, an idea that I call the Buena VistaSensing Club hypothesis. Given prior dis-cussions of the relationship between plan-ning and consciousness (Bridgeman, 1992,20°3), the Buena Vista Sensing Club hypoth-

esis may also be a useful step in articulatingan empirical approach to the evolution ofconsciousness.

2.1. Morphological Computation

2.1.1. BAT EARSBats have been a staple model system ofneuroethology since Galambos and Griffin(1940) first elucidated the role of self-generated acoustic emissions in their noc-turnal hunting behavior. In the interveningyears, a host of laboratories have investi-gated how bats are able to perform theirhigh-precision, high-speed prey-capturemaneuvers in total darkness by using acous-tic pulses, emitted from either the nose ormouth. The bat is a very successful exampleof an active-sensing system (eight hundredecholocating species worldwide), in which

MALCOLM A. MACIVER

an animal perceives the world via a self-generated signal source (Nelson & MacIver,2006). From a signals perspective this is aremarkable feat, because all active-sensinganimals must overcome spherical spread-ing (r2 ) losses as a signal propagates awayfrom them, as well as spherical spreadinglosses from the target back to the source,resulting in quartic attenuation of the signalwith distance. Thus, to double their sensingrange, active-sensing animals need to gen-erate a signal at least sixteen times morepowerful (Nelson & MacIver, 2006). As theyhunt for prey, bats emit pulses of acous-tic energy at frequencies of thirty kilohertzand above. The spectral composition of thesignal varies greatly with species and habi-tat (Schnitzler et a1., 2003)' For example,bats that hunt in cluttered environmentsgenerate a call that has two components: aconstant frequency (CF) portion followedby a downward frequency-modulated (FM)sweep. The CF portion of the call givesthe bats better distance acuity, whereas theFM portion gives them multispectral cuesconcerning fine details of target shape andmovement (Suga, 1990). During the finalphase of a prey-capture sequence, calledthe "terminal buzz" phase, the bat rapidlyincreases the pulse rate (Ghose & Moss,2003; Kalko, 1995) and switches exclusivelyto FM sweeps.The bat is able to detect its horizontal

angle (left-right bearing, or azimuth) to tar-gets by analyzing both intensity and timedifferences in the sounds arriving at the twoears: if the target is to the left, the sound willarrive slightly earlier and with more intensityat the left ear than the right. However, anobject's vertical position is not detectable inthis manner. Instead, the intricate shapes ofthe bat's ear (pinnae) and tragi (Figure 26.3)provide cues to vertical elevation 0Notton,Haresign, & Simmons, 1995; Wotton & Sim-mons, 2000). Returning sonar cries followdifferent pathways through the pinna-traguscomplex according to their angle of entry,inducing spectral cues that vary systemati-cally with the elevation angle. The bat canthen simply listen to these spectral cues todetect the elevation of the target. The con-

formation of skin and supporting tissue ofthe ear in the bat forms a computationaldevice that solves a key problem in the local-ization of prey in three-dimensional space.

2.1.2 THE GEOMETRY OF THE FLY EYEMovement through space creates a patternof visual information called "optic flow." Asyou move forward through space, things tothe side seem to move backward. The rate atwhich they move backward is a function ofyour distance to them, the horizontal angle,and your velocity: things looming directlyahead seem to hardly move at all. If you arerolling around the axis of forward motion,rather than translating (suppose you are fly-ing in an ultralight airplane near the groundand a gust ofwind hits so that one wing goesup and the other down), a different optic-flow pattern occurs. Flies detect self-motionto help stabilize flight, and neuroethologicalevidence is emerging that the geometry andwiring of the photoreceptors on the eye ofthe fly make computing optic flow a trivialproblem (Egelhaafet a1., 2002). A key findinghas been that the orientations of the rows ofphotoreceptors along which the optic flowis computed, a function of the eye's geom-etry, coincide with the preferred directionsof the neuron that those sensors connect to(Figure 26.4). A neuron dedicated to detect-ing a rolling motion of the fly connects toa row of sensors, ommatidia, that lie paral-lel to the optic-flow pattern that occurs onthe eye when the fly rolls. Activation of thatneuron would then be a reliable indicatorthat the corresponding optic flow, and thusrolling self-motion, is occurring.The examples of bat ears and fly eyes

show how the physical configuration ofthe body performs a sophisticated compu-tational role in the life of these animals.Other examples include spectral filteringthrough pigmented oil drops in the eyes ofbirds (Varela, Palacios, & Goldsmith, 1993),distinct filtering .properties of ampullaryversus tuberous electroreceptors in elec-tric fish because of the presence of tightlypacked skin cells in the lumen (Szabo,1974), phonotaxis in crickets through reso-nance in their forelimbs (Michelsen, 1998;

o 45 90 135 180ventral

Azimuth nFigure 26+ Computing optic flow through eye geometry.a. Self-motion generates optic flow over the eyes. Arrows on theleft plane represent the local motion vectors on the eye when theanimal rolls around its longitudinal body axis. The local responseproperties of a neuronal tangential cell, the VS6 cell, are adapted todetect this particular self-rotation. It is assumed that with its largedendrite, this cell integrates signals from local input elements whosepreferred directions (arrows on second-from-left plane) correspondto the direction of local motion vectors in roll-induced optic flow.b. Head of a female blowfly. White lines over the right eye indicatethe course of ommatidial rows in the eye lattice. c. Organization ofthe receptive field of a VS6 cell. Orientation and length of arrows atdifferent angular positions indicate the neuron's local preferreddirection and motion sensitivity in the right visual hemisphere. 0°azimuth and 0° elevation corresponds to the point directly in frontof the animal. Lines in the upper-left quadrant indicate the courseof ommatidial rows, which are oriented vertically in the equatorialregion of the eye (v-row). The direction of visual motion is thoughtto be analyzed mainly by interactions between ommatidia along therows in the hexagonal eye lattice (cf. orientation of rows andarrows). In the dorsofrontal eye region, the course of the v-rowsstrongly shifts toward a horizontal orientation. This change inorientation is reflected by the change in local preferred directions ofVS6 cells in corresponding regions of its receptive field. Textverbatim from Egelhaaf et a!. (2002). Reproduced with permission.

dorsal

b

adaptive behavior occur} in part} outsideof the nervous system} a useful prelimi-nary step toward the argument that cog-nition can depend on processes outside ofthe body. An important issue holding upacceptance of these points is how to relatethe somewhat incommensurate form of ana-log computation apparent in these cases to

NEUROETHOLOGY

cTangential cell

YS6

. . . . . .""" '

' .. ;

" ,', .. . . /---'

Directions ofoptic flow vectors

during roll

".. '.', / I Preferred directions of. local input elements

a

Michelsen, Popov} & Lewis, 1994), and therole of cochlear micromechanics in hear-ing (Gummer} Hemmert, & Zenner} 1996;Russell & Kossl, 1999)' In these cases, asignificant amount of signal processing hasbeen completed prior to entry of the sig-nal into the nervous system. These examplesshow that key computations subserving

49° MALCOLM A. MACIVER

the more mature notions of computationand algorithm complexity with digital rep-resentations (for an insightful discussion ofthe distinction between analog and digital,see Haugeland, 1981). This question, howcomputability over discrete spaces relates tocomputability over continuous spaces suchas real numbers, has become a topic of inter-est over the past few years, and metrics suchas the log of condition in the case of linearsystems have been proposed (see reviews byBlum, 2004; Braverman & Cook, 2006).

2.1.3 BILATERAL SYMMETRY IN ANIMALBODY PLANSThe bat and fly examples concern exploita-tion of structure in the sensory ecologyof those animals. How do animals exploitstructure in their mechanical ecology? Adeep pattern connecting sensory processingand locomotion can be read in the bilat-eral symmetry of animals. All animals butsponges, jellyfish, comb jellies, and placo-zoans are bilaterally symmetrical. The bila-terian animal body plan, which appearedcontemporaneously with the appearance ofmulticellular animals between 0.5 and 1. 5bil-lion years ago, often includes high forwardmobility along the midline axis (Grabowsky,1994) and is closely coupled to cephaliza-tion, an adaptation that complements suchmobility with a clustering of sensory organsaround the anterior end of the organism. Allvertebrates and most other highly mobileanimals such as insects feature this neurome-chanical complementarity. With this neu-romechanical template (Full & Koditschek,1999) came the active feeding behaviors andagile locomotion that correlate with the evo-lution of advanced nervous systems neededto control these behaviors (Conway, 1998;Dewel, 2000; Koob & Long, 2000; North-cutt, 2002; Paulin, 2005)' Iwill return to theseissues below when I discuss why the abilityto sense objects distant from the body origi-nated.Consider what would happen during for-

ward movement in a fluid were it notfor symmetry around the midsagittal planealong which animals propel themselves:there would be an imbalance of drag forces

on the body, making the animal yaw (turnleft or right) from the trajectory that isthe shortest distance between two points.According to one theory of its origins, bilat-eral symmetry arose when a jellyfishlikeancestral form migrated from a midwa-ter existence to crawling along the seafloor(Trochaea theory, Nielsen, 2001). Even whenthey are not moving on an aquatic or terres-trial surface, however, given their presencein volumes of water or air close to surfaces,all animals are more than two-dimensionalcreatures but not fully three-dimensionalones - even birds and fish typically movefar less up and down than horizontally. Astwo-and-a-half-dimensional creatures, ani-mals can exploit their bilaterally symmet-rical sensory structures, such as ears, forcontrol of horizontal bearing by simplecomparisons between them, as in the bat(termed tropotaxis, and achieved by com-missural fibers connecting bilaterally pairedsensory nuclei in the brain; for more details,see Braitenberg, 1965, 1984; Hinde, 1970;MacIver et al., 2004)' Thus, fundamentalneeds of efficient locomotion, and sim-ple sensor-based control algorithms for theimportant decisions ofleftward versus right-ward movement via bilaterally paired sen-sory organs, appear to have been entrenchedin the structure of animals since the dawn oftheir multicellular origins.

2.1.4 THE PASSIVE WALKERAnother observation concerning the impor-tance of mechanical structure in subserv-ing behavior comes from outside of neu-roethology, in the field of robotics. Overthe past several years, building on the workof McGeer (1992), Andy Ruina and oth-ers have been working on passive walk-ers. These robots, designed to be modelsof human bipedal walking, are able to walkwith no energy other than the small amountimparted by an inclined plane, and nothingother than rigid links and joints. More recentversions use a very small amount of energy,similar to the amount used by humans,to walk on flat surfaces (Collins, Ruina,Tedrake, & Wisse, 2005). The implicationof this work is that much of the efficiency

NEUROETHOLOGY 491

.'

prey.-.

;'

, '6/ /b3 senlry volumep;edator" )

stopping motor volume

/

derived commonly observed gaits in bipeds(Srinivasan & Ruina, 2006) and movementsin fish (MacIver et a1., 2004)'

2.1.5 THE BONE-BRAIN CONTINUUMA general pattern emerging from the workabove is that regularities existing over dis-parate timescales in an animal's environ-ment are absorbed by different systems ofthe body. The longest timescale regularities,such as the need for a balance of forces alongthe axis of travel to move in straight lines,are encapsulated at the level ofstructural tis-sues (bones, cartilage, chitin) .in a bilaterally

b1

/'/

b

lines of trees ,and bushes \

sensory volume

stopping motor volume

prey

Figure 26.5. From reactivity with a mala vista to planning with abuena vista. Control and planning in prey-capture behavior as afunction of the ratio between the sensory volume (SV) and thestopping motor volume (MVstop) for two fictive animals: one (left)with the near unity ratio characteristic ofmany passive sensinganimals that have poor acuity and active-sensing animals andanother (right) with a large ratio to illustrate the rarer situation oflong-range passive-sensing systems such as vision. a. Withnear-unity SV:MVstop ratios, search proceeds in a raster-sean-likefashion through the environment. If a prey is close enough to bewithin one of these search tracks, it is detected and possiblycaptured. b. With large SV:MVstop ratios, there is the pOSSibilitythat multiple trajectories to a detected prey (dashed lines) can beassessed prior to action. After assessing multiple paths, one path ischosen (b1) that is longer than a path that may disclose the positionof the predator to the prey too early (b2) or result in reaching anuntraversable obstacle (b3).

a

ofhuman movement may arise through hav-ing a skeletal structure and mass distributionthat makes walking as energetically favor-able to the body as swinging is to a pendu-lum. Such efficiency is enormously impor-tant: if we were as inefficient as the Hondabipedal robot Asimo, which requires at leastten times more energy per unit distance andweight (Collins et a1., 2005), then with afull day's walking we would need to eatten times more food than we typically eat.Further evidence of mechanical efficiencycomes from computational studies using thetechnique of optimal control to find move-ments that minimize energy. These have

MALCOLM A. MACIVER

symmetrical body plan. The maintenancecosts of these tissues are the lowest of anybody material. The shortest timescale reg-ularities, such as the current state of self-motion, are encapsulated at both the struc-tural level of sensor geometry and throughenergetically expensive neuronal process-ing, tissue which requires forty times morepower per unit mass than bone (Martin &Fuhrman, 1955). The computations that ananimal needs to move through space arespread across these systems. At this level ofdescription, there is no basis for an invidiousdistinction between bone and brain.

2.2 The Buena Vista Sensing Club

The Buena Vista Sensing Club hypothesissuggests that the expansion of the range withwhich animals can monitor external space,relative to their usual velocity, has beenone - perhaps the dominant - driving forcefor the evolution of the ability to plan. Inmost environments, distant goals cannot beeffectively reached by single behaviors butinstead require the sequencing of multiplebehaviors. Although the sequencing couldbe purely reactive, it is clear that multipleapproaches are possible when our percep-tual world extends to a large expanse ofheterogeneous space (Figure 26'5b); thus, atthe very least, there is a basis for selectionpressures that would lead to a capacity forevaluating multiple possible trajectories to agoal. The number of animals with the req-uisite sensory-mechanical balance conserva-tively includes cephalopods (squids, cuttle-fish, octopus), raptors such as hawks andeagles, and many of the mammals. Beforeproceeding, I would like to address the ques-tion of why sensation jumped beyond theboundaries of animals in the first instance.

2.2.1 A BRIEF HISTORY OF TELECEPTIVESENSATIONIn his Opticks, Newton remarked, "Infinitespace is the sensorium of the deity." In whatways is finite space the sensorium of lessercreatures? First, a bit of evolutionary back-ground. Between 700 million years ago, thedate of the earliest trace fossils ofmulticellu-

lar animals, and 1.6 billion years ago, the esti-mated time of the last common ancestor ofplants and animals, all animals were unicel-lular and thus lacking a nervous system (Car-roll, 2001; Meyerowitz, 2002). It seems likelythat the nervous system is a solution to theproblem of controlling the body when it iscomposed ofmore than a few cells (Nielsen,20m), at which point diffusion breaksdown as an effective communication system.Much of the nervous system is concernedwith problems that arise from being mobile,and it appears that its evolution was greatlyaccelerated after animals discovered anappetite for eating other animals (Conway-Morris, 1998; Northcutt, 2002; Paulin, 2005).Consider the example of the tunicate, whichhas a mobile, bilaterally symmetrical lar-val stage, and a stationary (fixed to a sub-strate), asymmetrical adult phase: once ithas reached its final resting spot, the tunicatedigests much ofits nervous system. Whetheror not ignorance is bliss, for a sessile crea-ture a brainless existence at least reduces theamount of grub it has to come bylLet us consider the problem of sensory-

signal-based movement guidance in somedetail, because it appears central to the gen-esis of nervous systems. Our ancestors areunicellular animals for which, by dint oftheir small size, inertial forces on the bodyare dominated by drag forces (the ratio ofthese two is quantified by the Reynoldsnumber). The simple consequence of thisis that as soon as you stop generating forcesto move, you will stop. Think of walkingon normal ground versus skating on ice. Inthe former, as soon as you stop generat-ing forces through your contact with theground, you stop moving; in the latter, evenafter you stop pushing off the ice, you arestill moving. In water, our ancestral envi-ronment, animals are in the viscous regimewhen they are below around a millime-ter in size. As an animal grows larger, itenters the inertial regime, and now in theabsence of active braking forces the ani-mal will coast along through space for sometime after cessation offorce production. Therelationship between control signals, such asthose required to whip a flagellum to move

-

NEUROETHOLOGY

forward, and the point in space that the ani-mal needs to reach, perhaps some tasty bac-terium, is not straightforward in the inertialregime. This is because the animal cannotsimply halt motion at the point of contactwith the bacterium but needs to performstate estimation to predict how soon beforethe bacterium is reached it will need to shutoff the generation of force as a function ofits current dynamics (Paulin, 2005)' Con-tact sensors such as mechanoreceptors arenot sufficient: you need a teleceptive sen-sory system - one that can detect targetssome distance away from your body with-out contact. There are several of these, mostobviously the visual system, but also audi-tory, chemosensory, electrosensory, and themechanosensory lateral line, which detectsflow disturbances in water at a distance.For a given object, the maximal distanceat which it can be detected using one ofthese teleceptive systems - as one movesthe object in all directions around the body-forms a surface we will refer to as the sensoryvolume (SV).

2.2.2 TELECEPTIVE SENSATIONAND CONTROLWhat determines how far away from thebody, and in what directions, the sensoryvolume should extend to allow for effectivecontrol of the body in space? Consider thefollowing scenario: you are driving along inthe fog, able to only see a short distance infront of the car. Suddenly, a huge lumber-ing moose appears through the fog, stand-ing in the roadway. You now have severaloptions: step on the brakes, swerve, run intothe moose, or some combination of these.Unfortunately, any action you take willbe at least two hundred milliseconds afterthe moose-related sensory signals hit yourretina, because ofconduction and processingdelays between the surface of your retinaand the contraction ofmuscles in your loweror upper limbs. Now, if you see the moosethree meters in front of the car, and youare going fifty kilometers per hour (four-teen meters per second), by the time youstep on the brake, the moose is alreadygoing through your windshield. If we allow

493

about five hundred milliseconds for an eva-sive action such as braking or swerving tooccur after you have initiated it (althoughactually it will be a function of, e.g., yourvelocity, mass, friction of the roadway-tireinterface, braking power, speed of musclecontraction), that means you need to sensethe moose around seven hundred millisec-onds before contact, or ten or more metersaway if you are going fifty kilometers perhour. In short, the neuromotor delay timeplus the action time determine an effec-tive horizon of reactivity; if you sense thingsinside of that horizon, you are powerless; ifyou sense things outside of the horizon, youwill at least have time for the most basic ofactions. This example gives a sense of thedynamic considerations involved in deter-mining how far the sensory volume shouldextend in the direction of movement. Asyou extend your sensory volume beyond thehorizon of reactivity, you allow for morethan simple reactive control strategies, suchas braking or swerving. With a buena vista,you can look far ahead and execute long-duration plans such as multiple lane changesprior to an exit. However, note that theseconsiderations only apply to less predictablefeatures of the space - a where,as Haugeland (1998) nicely puts it, "percep-tion is cheap, representation expensive" (p.219)' In less dynamic contexts, such as long-range navigation, and where perception iscostly, guiding movement through internal-ized spatial maps may be more effective.If, for example, that moose on the road infront ofyou is actually stuffed, and has beeninconveniently installed in the center of theroad, you could eventually learn to avoidit early on, on the basis of your recognition,say, that it is just past the hairpin turn beforeCrazy Bob's Taxidermy and Pedicure. Thisrepresents the internalization of space in aform that, other than issues of changes inthe space and the need to localize your posi-tion in the space, may be interchangeablewith sensation. I will return to this point todiscuss reports that active-sensing animalspossess highly accurate spatial maps.Whereas we have the concept of sensing

range for discussing how the sensory volume

MALCOLM A. MACIVER

f •

494

relates to control, there is no analogous con-cept for movement. In control theory, a dis-cipline of engineering, where something canmove over a given time span is called the"small-time reachable set," and I proposethat this concept provides a usable motorsystem analogue to sensing range. To defineit, we begin with a mechanical system char-acterized by a set of time-varying controlinputs (e.g., for a car it could be rear-wheelrotational position and front-wheel steeringangle). For this mechanical system one canestimate the small-time reachable set to bethe region of space that the mechanical sys-tem can reach over a given time intervalfor all feasible control inputs (i.e., inputsthat do not exceed the capacity of the sys-tem, such as a ninety-degree turning angleor acceleration beyond the power of theengine). The original work in this area con-cerns reachable sets of a kinematic car (e.g.,Vendittelli, Laumond, & Nissoux, 1999), butmore recent work treats the computationof reachable sets for continuous dynami-cal systems (Mitchell, Bayen, & Tomlin,2°°5),With the concept of the small-time

reachable set, along with the notion of thesensory volume, I can address how the rela-tive sizes ofmovement and sensing volumesrelate to behavioral control. I will do thisin the context of one popular model systemfor the study of sensory processing in ani-mals, the weakly electric fish.

2.2.3 MOVEMENT AND SENSING SPACESIN WEAKLY ELECTRIC FISHFor studies of the weakly electric fishApteronotus albifrons, the black-ghost knifefish, my colleagues and I have applied theconcept of the small-time reachable set(hereafter termed the motor volume, MY)to determine the extent of the overlapbetween the motor volume and the sensoryvolume for detection and capture of prey(Snyder, Burdick, Nelson, & MacIver, 2007).As an object enters the weak electric fieldthat the fish continually emits, distortionsin the electric field are picked up by aroundfourteen thousand electroreceptors coveringthe entire body surface. These distortions

in the field are analyzed and used to directsubsequent behavior. Using the electric fieldand sensors as a teleceptive active-sensingsystem, the fish is able to hunt at night inthe muddy rivers of the Amazon basin.We quantified the three-dimensional

shape and size of the electric fish's preySV using a combination of behavioral andcomputational techniques that allowed usto estimate when the live prey could bedetected (Snyder et a1., 2007). The omni-directional SV for prey is shown in Figure26.6. The MV varies as a function of thetime interval being examined, the fish's ini-tial state, and the control inputs of thefish's musculoskeletal system over the timeinterval being considered. Because we donot have access to these control inputs, weestimated the MV empirically by analyz-ing motion capture data of these fish hunt-ing for prey. The strategy we used was tolook at all body displacements that occurredover a given time interval across multipletrials. We quantified the MV by placing asurface over the maximal displacements inall directions; thus, for each time interval,we obtained a particular MY. A fascinat-ing and unexpected finding was that, similarto the SV, the MV is also omnidirectional,a testament to the remarkable morphologyandmaneuverability of these animals. Figure26.6 shows the MV for three different timeintervals, showing that it becomes largerand changes shape as the interval increasesfrom over one hundred to seven hundredmilliseconds.We can now quantify how closely

matched these two spaces are by coming upwith a convenient measure of this match;we use the intersection of the two volumesdivided by their union. When we exam-ine this measure versus the time interval ofmovement, we find that the match is maxi-mal at a time interval of about 432 millisec-onds. The importance of this time interval isthat it is close to the sum of the neuromotordelay and stopping times for the fish; it takesone hundred milliseconds for a detectablesignal indicating prey to reach the brain andproduce a behavioral reaction, and anothertwo hundred milliseconds are .required for

NEUROETHOLOGY 495

117 ms

with coming to a stop as an animal's stop-ping motor volume (MVstop).When an animal needs to emit sixteen

times more power to double its sensingrange, we expect selection pressure againstmaking the SV any larger than it needs to be.Here is where the great advantage ofpassive-sensing systems, such as vision, becomesapparent: now the only cost associated withhaving a longer sensing range is havinga larger eyeball and associated visual sig-nal processing circuitry (Brooke! Hanley, &Laughlin, 1999; Land & Nilsson! 2002). Withthe resolution of our high-acuity visual sys-tems, about one-sixtieth of a degree underideal conditions, we can resolve a thirty-centimeter-long rabbit at one kilometer - a

sensory volume

Figure 26.6. Sensory volume and motor volumes of a weaklyelectric fish. The prey SV (for a typical prey, the water fleaDaphnia magna) and MVs of a weakly electric fish as a functionof the indicated movement time. Adapted from Snyder et al.,( 20°7)'

the animal to come to a halt from stan-dard hunting velocities and capture the prey.In other words, in an active-sensing animal,which has to invest metabolic energy to pro-duce a sensing field, just enough energy isemitted for the animal to be able to detectthe prey far enough out to stop. (Like manyanimals, these electric fish simplify the con-trol problem ofengulfing small prey by beingnearly stationary at the point of ingestion.)From an analysis of previously published

data on bat prey capture we find evidenceof a similar pattern: the echolocation SV(Figure 26.1a) is close to the MV at the sumof the neuromotor delay and the stoppingtime (Snyder et a1.! 2007). For convenience,I will refer to the particular MV associated

MALCOLM A. MACIVER

distance that takes us more than ten min-utes to walk at a good clip and about fourminutes at a fast run. At night, and whenour mechanical abilities are augmented bya vehicle, the relationship between our SVand MVstop is not necessarily intuitive: weare strapping on a massive, fast mechanismthat expands our MVstop , while operatingwith low light levels, which shrinks ourSV. The two factors combined with ourconsiderable neuromotor latency result insurprisingly slow recommended night driv-ing speeds from highway safety agencies(less than fifty kilometers per hour whenusing low-beam headlights; U.S. NationalTraffic Highway Safety Administration,2004).

2.2.4 CONTROL AND PLANNINGHow does this relate to control and plan-ning? If the ratio of the SV to MVstop isnear one (Figure 26.5a), the set of behav-ioral options with respect to something justsensed is very small, limited to stopping,turning, and other behaviors that have sim-ple relationships between sensory input andneuromotor output. We refer to these sim-ple behaviors as "reactive" because thereis only time for simple reflexive reactions.With simple mappings between inputs andmovement, animals can embed control lawsin parts of the nervous system like the spinalcord and hindbrain that feature minimalprocessing and conduction delays. As weincrease the ratio of SV to MVstop , an ani-mal has the capacity to examine a largerspace beyond that which it will immedi-ately move several behavioral cycles into thefuture, where a behavioral cycle is the neu-romotor delay time plus the time it takesto perform a simple action like stopping.The first consequence is that when targets offuture behavior come into range, the animalhas many options, and dynamics and neu-romotor delays do not dominate potentialresponse modes (Figure 26'5b). For example,an animal could sequence multiple behav-iors to reach a goal, such as trotting over toa barrier, jumping over it, and then crouch-ing under an obstruction. An animal with a

low SV:MVstop ratio has little ability to eval-uate different trajectories toward or awayfrom the sensed object because of time andspace constraints. With a large ratio, therewill be multiple feasible trajectories to adistant goal. If these different trajectoriesvary in their likelihood of success, we wouldexpect selection pressure to favor animalsthat are able to evaluate these different pos-sibilities and then select the one most likelyto lead to success.Returning to the scene of Crazy Bob's

Taxidermy and Pedicure, it is important tonote that the preceding claims about controland planning are qualified by "with respectto something sensed." This is because weknow that the ability to navigate throughforaging areas far larger than the SV andreturn home is common among animals,whether they possess short- or long-rangesensing systems. It appears that many ani-mals develop a cognitive map of the largerterritory over which they live, and they canindex their position in this space either byusing local landmarks or through path inte-gration, where current position is estimatedby updating some initially known positionwith each subsequent movement the ani-mal makes (Hafting, Fyhn, Molden, Moser,& Moser, 2005)' Provided that the featuresof interest in the space are not significantlyaltered between visits, it is quite conceivablethat such animals could plan to revisit thesefeatures while relying only on path integra-tion or on sensing of nearby landmarks.In this way, a representation of space plus

path integration or localization may be agood proxy for sensation. However, thereare clear differences. Each unit of space con-tains potential harms and benefits for a typ-ical animal. These fall into two categories:those items that are stable enough throughtime that their presence or absence can beinternalized into an infrequently updatedmap of space, and those items that are not.In the former category are things such asenvironmental obstructions, the location ofhome, the level of the tide, where food hasbeen cached for use during the winter, thelocation of flowering plants, and so forth.

NEUROETHOLOGY

Into the latter category is the presentlocation of predators and prey and cur-rent weather conditions. Although there areclear advantages to being able to plan overstable features of the environment, planningover the SV is critical for predation andavoiding being preyed on. Because the shiftto a mobile-predator lifestyle in the earlyCambrian (more than five hundred millionyears ago) likely led to the innovation of thevertebrate head and related sensory struc-tures from headless ancestors (Northcutt &Gans, 1983), any capacity that affects preda-tory ability has the potential to be a source ofsignificant selection pressure on the nervoussystem of animals.It is nonetheless interesting to consider

some of the complex relationships amongsensing range, rate at which the habitatchanges over time, size of foraging area,animal speed, and accuracy of landmark-based guidance systems. For example, batsappear to prefer linear landmark featuresbetween roost and foraging area (Schnit-zler et al., 2003)' This preference probablyfacilitates their observed reduced relianceon echolocation along bat flyways (preferredpaths from roost to hunting grounds), as wellas allowing a cruder spatial map (e.g., lessfrequent location updates) than would benecessitated if the bat were to follow a morefractal landmark structure at their high fly-ing speed. Active sensing animals in generalseem to have particularly accurate spatialmaps (on electric fish, see Cain, 1995; Cain,Gerin, & Moller, 1994; on rodents, see Haft-ing et al., 2005; O'Keefe & Burgess, 2005;on bats, see Schnitzler et al., 2003)' Clearly,however, there are many different waysthat space can be internalized. For exam-ple, internalization could be as rudimentaryas following a trail of chemical laid downby a fellow traveler, such as occurs withants, or as complicated as the hippocampus-dependent distributed spatial-cue bindingfound in mammals (Cohen & Eichenbaum,1993)'A further difference between planning

over sensed space and internalized space isthat the latter may place significant demands

497

on structures such as the hippocampalformation that appears to be central to spa-tial cognition (Cohen & Eichenbaum, 1993;O'Keefe &Nadel, 1978; Hafting et al., 2005).An example of this is the seasonal expan-sion and shrinking of related brain struc-tures in food-caching birds (Krebs, Sherry,Healy, Perry, & Vaccarino, 1989; Sherry,Vaccarino, Buckenham, &Herz, 1989; Smul-ders, Sasson, & Devoogd, 1995)' The shrink-ing and expansion ofbrain tissue with spatialmemory load in these animals suggests thatmemory can be quite costly (Dukas, 1999)'Quantification of this cost would allow usto assess the trade-off between maintaininga rapidly updated representation of spacethrough sensation, with the associated costsof sensing (Laughlin, 2001), and a slowlyupdated representation that requires pathintegration or localization with the associ-ated costs of spatial memory.Ultimately, the tipping point between the

effectiveness of single control laws versusplanning (be it over internalized space orthrough sensing) may have to do with howdensely occupied space is with behaviorallyrelevant contingencies, relative to move-ment speed and to the size of the MVstop .This is similar to Levins's (1968) notion ofenvironmental grain. Living in a very sparseenvironment and possessing long-range sens-ing systems, long reaction times, and a largeamount ofinertia (being massive andJor fast)is similar to living in a cluttered environ-ment and having short-range sensing sys-tems, short reaction times, and low inertia.The rain forest is not equivalent to a bar-ren desert under equal sensory and move-ment conditions, but if the clutter of therainforest contracts the SV or the desertincreases speed, they could possess similarplanning loads in this framework. In birds,it has been shown that flight speed increaseswith (body mass)o.167, and visual resolutionincreases with (flight speed)L33, so largerbirds with more inertia resolve objects atlonger times to contact than do smaller birds(Brooke et al., 1999)' In bats, prey detectionrange is matched to the wingbeat interval,which in turn has a power law relationship

MALCOLM A. MACIVER

to body mass and flight speed (Holderied &von Helversen, 2003).

2.2.5 PUITING BUENA VISTA TO THE TESTTo test the Buena Vista Sensing Clubhypothesis, ideally we would start withbehavioral correlates of planning in a widerange of animals and relate them to theirsensory and motor capacities. Unfortu-nately, adequate behavioral data on planningbehavior are not available. Quantitative dataon the motor and sensing spaces of a vastarray of animals other than more modernvertebrates are also lacking. Cognitive neu-roscientists, however, have been investigat-ing the neural locus of planning in humansfor some time. Thus, in lieu of the missingbehavioral data, we will consider some ofthis evidence.In humans, cognitive neuroscience has

shown that the prefrontal cortex is impor-tant for planning (Damasio, 1985). Withinprimates, we know that humans have nearlydoubled the volume of Brodmann's area 10,a prefrontal cortical area considered impor-tant for planning, over the closest nonhumanprimate (Semendeferi, Armstrong, Schle-icher, Zilles, & van Hoesen, 2001), whereashuman visual acuity is not significantly bet-ter than that of other primates (Ross, 2000).This may be an exception to the BuenaVista Sensing Club hypothesis. However,the ability to plan without the help of sym-bolic methods such as language may havereached saturation in early primates. It ispossible that with language and other sym-bol systems, hominids took planning to anew level. Although language is thought tobe a more recent innovation than is com-patible with increased frontal lobe volume,given recent evidence concerning the audi-tory capacity ofearly hominids, language usemay be quite ancient (Martinez et a1., 2004)'Symbolic approaches can effectively extenda perceptual system to encompass an in-definite amount of space for planningthrough the reports from fellow symbolusers; one example of this is the bee's waggledance, which allows individual bees to com-municate the location and richness of a for-aging patch far beyond the SV ofbees at the

hive. Given these considerations, compar-ing nonhuman primates, which as a grouphave larger eyes than other mammals (Ross,2000), to other vertebrates may be moreinformative. Along these lines, it is knownthat mammals (which appear to make upthe bulk of the members of the BuenaVista Sensing Club) have greatly increasedthe complexity of the forebrain overother vertebrates (Figure 26.7; Striedter,2005)'One further point should be made to but-

tress the Buena Vista Sensing Club againstcomplaints from members of the Mala VistaSensing Club - all those animals with punysensing ranges. They could argue that if any-one should have been pressured into havingan ability to plan, it should have been them,because they would benefit even more fromthis ability than a Buena Vistite. There arethree primary ways in which selection pres-sure to plan can be manifested. One is toevolve long-range sensing abilities, thus join-ing the Buena Vista Sensing Club. Anotheris through symbol use, such as language orwaggle dances. The third is through an inter-nal map plus path integration to determinewhere one is in that map, as discussed above.The first, joining the Buena Vista SensingClub, is inapplicable, as we are address-ing members of the Mala Vista Club. Wewill not consider the second, as symbolicapproaches are rare. The third approachis possible, but we return to our previouspoint, that a capacity to plan over internal-ized space will not help with predation andavoiding being preyed on, both ofwhich aresignificant sources of selection pressure onthe nervous system. Whether other sourcesof selection pressure are sufficient to lead toplanning is unclear. The presence of accu-rate spatial maps, and perhaps planning, inactive-sensing animals would seem to sug-gest the answer is yes. However, two ofthe active-sensing modalities, echolocationand electrolocation, are in animals that haverelatively recent, non-active-sensing ances-tors. The ability to internalize large mapsmay be a holdover from long-range passivesensing habits and the exploitation of corre-spondingly large foraging areas .- an echo of

257

NEUROETHOLOGY 499

are encapsulated in the animal at a hostof levels from structural tissues to neuralresponses, according to the temporal band-width of these regularities and associatedenergy trade-offs. For the second step, I elu-cidated how work on one model system inneuroethology, the weakly electric fish, maygive insight into relationships among con-trol, planning, and the ratio between the SVand MV5top . I put forth the hypothesis thatsensing beyond the MVstop may be central tothe evolution ofplanning. If true, the BuenaVista Sensing Club hypothesis naturalizes aformerly largely human faculty in a way thatmakes it approachable in more experimen-tally accessible animals.Adams and Aizawa (this volume) raise

the issue of the motivation for an exter-nalist redefinition of cognition. As part ofthe effort to understand the intelligenceof nonhuman animals, for neuroetholo-gists the redefinition is driven by pragmaticneeds. Craniocentrism simply does notwork for understanding the kinds of quitesophisticated behaviors that neuroetholo-gists are working on. As suggested by thethree virtual-world examples, the instancesof morphological computation, and therelationships among behavioral control,

Figure 26'7' Toward a comparative biology of planning.Phylogenetic tree depicting the relationships ofmajor vertebrategroups. Numbers across the top indicate the number of cell groupsexperts have described in the forebrains of representative species.The symbols indicate approximately where forebrain complexity islikely to have increased (squares) or decreased (circles). Modifiedwith permission from Striedter (2005).

Hagrishes Sharks Polyplerus lungfishes Frogs Birdslampreys Rays Teleosts Salamanders Reptiles Mammals

an condition when perception wascheaper.

3. Conclusion

I began with a precis of how neuroethologynavigates the sometimes contested realm ofits own externalist and internalist leanings.In excising a piece of the world with an ani-mal and coupling these together in closed-loop apparatuses, neuroethology recognizesthe unity of the external and internal inadaptive behavior. The virtual world exam-ples also show that in some cases we under-stand the relevant external factors needed toelicit natural behaviors, whereas the neuraland biomechanical underpinnings of thesebehaviors are still largely not understood;this in some measure supports an internalistbias in effort if not in philosophy.As the first of two steps from neu-

roethology to situated cognition, I showedhow present research on situated nervoussystems within neuroethology is indicatingthe important computational role of non-neuronal sensory and mechanical structuresin supporting adaptive behavior. Regular-ities in the world at multiple timescales

MALCOLM A. MACIVER

sensing, and mechanics I have discussed,they arise out of a tight interplay of body,brain, and environment, with deep ties toan animal's environment and evolutionaryhistory.I will end with some points about the

implications of the Buena Vista SensingClub hypothesis as it relates to extendedmind issues and consciousness. In Clark andChalmers's (1998) paper on the extendedmind, they give the example of Otto theAlzheimer's patient and Inga - whose mem-ory is unimpaired - determining the addressof a museum. They argue that the address ofthe museum in Otto's address book, alwayswith him, is functionally identical to Inga's(nonoccurrent) belief that the museum hassome particular address. I would argue thatthe realm of readily interrogable space deliv-ered by a long-range teleceptive sensorysystem can be equivalently thought of asan extended belief system; there as well,all potential subjects of perceptual fixationhave the status of some type of belief in theextended cognitive system. The idea of theenvironment serving as an external memorystory has been around since the 1970s, begin-ning with Dreyfus's work and later nicelyencapsulated by Brooks in his "let the worldbe its own bestmodel/l (see Noe, 2004, p. 234,m4). Specific proposals as to how percep-tual fixation via long-range sensing systems isimportant to cognition have been put forthby Dana Ballard and colleagues (Ballardet al., 1997)' The need for members of theBuena Vista Sensing Club to manipulate thisextended information space (LaValle, 2006)ofbeliefs to achieve distant goals would havegone quite beyond the capacity of reactivecontrol strategies.Bridgeman (1992) wrote, "Consciousness

is the operation ofthe plan-executing mech-anism, enabling behavior to be driven byplans rather than immediate environmen-tal contingencies./I Similar ideas were putforth by Humphrey (1992, p. 42). Conscious-ness, Prinz argues (this volume), is con-cerned with attention to intermediate-levelrepresentations that are useful for action.What I have presented here is a suggestionthat planning, and perhaps therefore con-

sciousness, was only necessary once percep-tual systems delivered choices at such a dis-tance that reactive (nonconscious) controlschemes for action were no longer advanta-geous. The lead-up to consciousness couldtherefore have been a gift of space wroughtby passive teleceptive sensation in a nichewhere such acuity paid fitness dividends.The need to sequence behaviors over thisspace would then have given rise to execu-tive control structures in the brain, includingworking memory and attention, for carryingout these sequences, in a case study of how"it is not the animal's brain that organizes itsworld, but the evolutionary ecology of theanimal that organizes its brain/l (Reed, 1996,p.69)·Neuroethology, with its comparative

approach and close attention to evolu-tionary, ecological, behavioral, and neuralaspects of animal life, may at first seem anunlikely contributor to the field of situatedcognition. However, as I hope the precedingexamples have suggested, it is well poised topush forward our understanding ofhow sim-ple and direct behavioral responses to sen-sory input can give way to abilities we morereadily recognize as cognitive.

Acknowledgments

I thank Andy Barto, Heather Eisthen, KevinLynch, Mark Nelson, and Rob Wilson for com-ments on earlier drafts, and Michael Dickinson,Mike Paulin, and Georg Striedter for useful dis-cussions. Thanks to RolfMiieller for images ofbatpinnae from microcomputed tomography. Thiswork was supported by National Science Foun-dation Grant No. IOB-oSI7683.

References

Assad, c., Rasnow, B., & Stoddard, P. K. (1999).Electric organ discharges and electric imagesduring electrolocation. Journalo/ExperimentalBiology, 202(10), 1185-1193.

Ballard, D. H., Hayhoe, M. M., Pook, P. K.,& Rao, R. P. N. (1997). Deictic codes forthe embodiment of cognition. Behavioral andBrain Sciences, 20, 723-767.

NEUROETHOLOGY

Barth, F. G., & Schmid, A. (Eds.). (2001). Ecology0/ sensing. Berlin: Springer.

Blum, L. (2004). Computing over the reals:Where Turing meets Newton. Notices 0/ theAMS, 51(9),1024-1034.

Boonman, A., &Jones, G. (2002). Intensity con-trol during target approach in echolocatingbats: Stereotypical sensori-motor behaviour inDaubenton's bats, Myotis daubentonii. Journal0/ Experimental Biology, 205(18), 2865-2874.

Braitenberg, V. (1965). Taxis, kinesis, and decus-sation. Progress in Brain Research, 17, 210-222.

Braitenberg, V. (1984). Vehicles: Experiments insynthetic psychology. Cambridge, MA: MITPress.

Braverman, M., & Cook, S. (2006). Computingover the reals: Foundations for scientific com-puting. Notices 0/ the AMS, 53(3), 318-329.

Bridgeman, B. (1992). On the evolution ofconsciousness and language. Psycoloquy, 3(15).http://www.cogsci.ecs.soton.ac.uklcgi/psyc/newpsy?p5, accessed May 17, 2008.

Bridgeman, B. (2003)' Psychology and evolution:The origins o/mind. Thousand Oaks, CA: Sage.

Brooke, M. D., Hanley, S., & Laughlin, S. B.(1999)' The scaling of eye size with body massin birds. Proceedings 0/ the Royal Society 0/Lon-don Series B, 266(417), 4°5-412.

Cain, P. (1995). Navigation in familiar envi-ronments by the weakly electric elephant-nose fish, Gnathonemus petersii L. (Mormyri-formes, Teleostei). Ethology, 99(4), 332-349.

Cain, P., Gerin, W., & Moller, P. (1994)'Short-range navigation of the weakly elec-tric fish, Gnathonemus petersii L. (Mormyri-dae, Teleostei), in novel and familiar environ-ments. Ethology, 96(1), 33-45.

Carew, T. 1. (2000). Behavioral neurobiology: Thecellular organization 0/ natural behavior. Sun-derland, MA: Sinauer.

Carroll, S. B. (2001). Chance and necessity: Theevolution of morphological complexity anddiversity. Nature, 4°9(6823), 1102-1109.

Chen, L., House, J. L., Krahe, R., & Nelson, M. E.(2005)' Modeling signal and background com-ponents of electrosensory scenes. Journal 0/Comparative Physiology A, 191(4), 331-345.

Chiel, H. J., & Beer, R. D. (1997)' The brain has abody: Adaptive behavior emerges from inter-actions of nervous system, body, and envi-ronment. Trends in Neurosciences, 20, 553-557·

Clark, A. (1997). Being there: Putting brain, body,and world together again. Cambridge, MA:MIT Press.

5°1

Clark, A., & Chalmers, D. (1998). The extendedmind. Analysis, 58, 7-19.

Cliff, D. (1995)' Neuroethology, computational.In M. A. Arbib (Ed.), The handbook 0/ braintheory and neural networks (pp. 626-630).Cambridge, MA: MIT Press.

Cohen, N. 1., & Eichenbaum, H. (1993). Memory,amnesia, and the hippocampal system. Cam-bridge, MA: MIT Press.

Collins, S., Ruina, A., Tedrake, R., & Wisse,M. (2005)' Efficient bipedal robots based onpassive-dynamic walkers. Science, 3°7(5712),1082-1085.

Conway-Morris, S. (1998). The crucible 0/ cre-ation: The Burgess Shale and the rise 0/animals.Oxford: Oxford University Press.

Dahmen, H. J. (1980). A simple apparatus toinvestigate the orientation of walking insects.Experientia, 36(6), 685-687.

Damasio, A. R. (1985). The frontal lobes. In K.Heilman & E. Valenstein (Eds.), Clinical neu-ropsychology (pp. 339-375). Oxford: OxfordUniversity Press.

Dewel, R. A. (2000). Colonial origin for Eumeta-zoa: Major morphological transitions and theorigin ofbilaterian complexity. Journalo/Mor-phology, 243(1), 35-74·

Dickinson, M. H. (1996). Unsteady mechanismsof force generation in aquatic and aerial loco-motion. American Zoologist, 36(6), 537-554.

Dukas, R. (1999)' Costs of memory: Ideas andpredictions. Journal 0/ Theoretical Biology,197(1),41-5°.

Dusenbery, D. B. (1992). Sensory ecology: Howorganisms acquire and respond to in/ormation.New York: W. H. Freeman.

Egelhaaf, M., Kern, R., Krapp, H. G., Kretzberg,J., Kurtz, R., &Warzecha, A. K. (2002). Neu-ral encoding of behaviorally relevant visual-motion information in the fly. Trends in Neu-rosciences, 25(2), 96-102.

Full, R. 1., & Koditschek, D. E. (1999). Templatesand anchors: Neuromechanical hypotheses oflegged locomotion on land. Journal 0/ Experi-mental Biology, 202(23), 3325-3332.

Full, R. 1., & Tu, M. S. (1991). Mechanics of arapid running insect: 2-legged, 4-legged and6-legged locomotion. Journal 0/ ExperimentalBiology, 156, 215-231.

Galambos, R., & Griffin, D. R. (1940). The super-sonic cries of bats. Anatomical Record, 78, 95.

Ghose, K., & Moss, C. F. (2003). The sonarbeam pattern of a flying bat as it tracks teth-ered insects. Journal 0/ the Acoustical Society 0/America, 114(2), 1120-1131.

502 MALCOLM A. MACIVER

Gibson, 1. 1. (1979)' The ecological approach tovisual perception. Mahwah, NJ: Lawrence Erl-baum.

G6tz, K. G., & Gambke, C. (1968). Zum bewe-gungssehen des Mehlkafers Tenebrio molitor.Kybernetik, 4, 225·

Grabowsky, G. L. (1994). Symmetry, locomotion,and the evolution of an anterior end: A lessonfrom sea-urchins. Evolution, 48(4), 1130-146.

Gummer, A. W., Hemmert, W., & Zenner, H. P.(1996). Resonant tectorial membrane motionin the inner ear: Its crucial role in frequencytuning. Proceedings of the National Academy ofSciences, 93(16), 8727-8732.

Hafting, T., Fyhn, M., Molden, S., Moser, M.B., & Moser, E. 1. (2005)' Microstructure of aspatial map in the entorhinal cortex. Nature,436(7052), 801-806.

Hartley, D. 1., & Suthers, R. A. (1989)' Thesound emission pattern of the echolocatingbat, Eptesicus fuscus. Journal of the AcousticalSociety ofAmerica, 85(3), 1348-1351.

Haugeland, 1. (1981). Analog and analog. philo-sophical Topics, 12, 213-226.

Haugeland, 1. (1998). Mind embodied andembedded. In J. Haugeland (Ed.), Havingthought: Essays in the metaphysics ofmind (pp.207-238). Cambridge, MA: Harvard Univer-sity Press.

Heiligenberg, W. (1991). Neural nets in electricfish. Cambridge, MA: MIT Press.

Hinde, R. A. (1970). Orientation. In Animalbehavior: A synthesis of ethology and com-parative psychology (pp. 46-192). New York:McGraw-Hill.

Holderied, M. W., & von Helversen, O. (2003).Echolocation range and wingbeat periodmatch in aerial-hawking bats. Proceedings ofthe Royal Society ofLondon Series B, 270(153°),2293-2299.

Holscher, c., Schnee, A., Dahmen, H., Setia, L.,& Mallot, H. A. (2005), Rats are able to navi-gate in virtual environments. Journal ofExper-imental Biology, 208(3), 561-569.

Hopkins, C. D. (2005)' Projects in neu-roethology. Retrieved September 18, 2005,from http://instruct1.cit.comell.edu/courses/

Hughes, H. C. (1999). Sensory exotica: A worldbeyond human experience. Cambridge, MA:MIT Press.

Humphrey, N. (1992). A history of the mind. NewYork: Simon & Schuster.

Hutchins, E. (1995). Cognition in the wild. Cam-bridge, MA: MIT Press.

Kalko, E. K. (1995)' Insect pursuit, prey captureand echolocation in pipistrelle bats (Microchi-roptera). Animal Behavior, 50, 861-880.

Koob, T. 1., & Long, 1. H. (2000). The vertebratebody axis: Evolution andmechanical function.American Zoologist, 40(1),1-18.

Kramer, E. (1975)' Orientation of the male silk-moth to the sex attractant bombykol. In D. A.Denton & J. P. Coghlan (Eds.), Olfaction andtaste (Vol. 5, pp. 329-355). New York: Aca-demic Press.

Krebs,1. R., Sherry, D. F., Healy, S. D., Perry, V.H., & Vaccarino, A. L. (1989)' Hippocampalspecialization of food-storing birds. Proceed-ings ofthe National Academy ofSciences, 86(4),1388- 1392.

Land, M. F., & Nilsson, D.-E. (2002). Animal eyes.New York: Oxford University Press.

Laughlin, S. B. (2001). The metabolic costof information: A fundamental factor invisual ecology. In F. G. Barth & A. Schmid(Eds.), Ecology ofsensing (pp. 169-185). Berlin:Springer.

LaValle, S. M. (2006). Planning algorithms. NewYork: Cambridge University Press.

Levins, R. (1968). Evolution in changing environ-ments: Some theoretical explorations. Princeton,NJ: Princeton University Press.

MacIver, M. A., & Nelson, M. E. (1996). Top-ics in neuroethology. Retrieved October 17,2005, from http://nelson.beckman.uiuc.edulcourses/neuroetholl

MacIver, M. A., Fontaine, E., & Burdick, 1. W.(2004). Designing future underwater vehicles:Principles and mechanisms of the weakly elec-tric fish. IEEE Journal of Oceanic Engineering,29(3), 651- 659.

Maglio, P., Matlock, T., Raphaely, D., Cher-nicky, B., & Kirsh, D. (1999)' Interactive skillin Scrabble. In Proceedings of the Twenty-firstAnnual Conference ofthe Cognitive Science Soci-ety. Mahwah, NJ: Lawrence Erlbaum.

Marler, P. (1991). Song-learning behavior: Theinterface with neuroethology. Trends in Neu-rosciences, 14(5), 19<)-206.

Martin, A. W., & Fuhrman, F. A. (1955)' The rela-tionship between summated tissue respirationand metabolic rate in the mouse and the dog.Physiological Zoology, 28, 18-34.

Martinez, I., Rosa, M., Arsuaga, 1. L., Jarabo, P.,Quam, R., Lorenzo, c., et al. (2004), Auditorycapacities in Middle Pleistocene humans fromthe Sierra de Atapuerca in Spain. Proceedingsof the National Academy of Sciences, 101(27),9976-9981.

NEUROETHOLOGY

McGeer, T. (1992). Principles ofwalking and run-ning. In R. M. Alexander (Ed.), Advances incomparative and environmental physiology 11:Mechanics of animal locomotion (pp. 113-139).Berlin: Springer-Verlag.

Meyerowitz, E. M. (2002). Plants compared toanimals: The broadest comparative study ofdevelopment. Science, 295(5559), 1482-1485.

Michelsen, A. (1998). The tuned cricket. News inPhysiological Sciences, 13(1), 32-38.

Michelsen, A, Popov, A. Y., & Lewis, B. (1994).Physics of directional hearing in the cricketGryllus bimaculatus. Journal of ComparativePhysiology A, 175(2), 153-164.

Miller, 1. A., & Surlykke, A. (2001). How someinsects detect and avoid being eaten by bats:Tactics and countertactics of prey and preda-tor. Bioscience, 51(7), 570-581.

Mitchell, 1. M., Bayen, A. M., & Tomlin, C.1. (2005)' A time-dependent Hamilton-Jacobiformulation of reachable sets for continuousdynamic games. IEEE Transactions on Auto-matic Control, 50 (7), 947-957.

Murlis, 1., Willis, M. A., & Carde, R. T. (2000).Spatial and temporal structures of pheromoneplumes in fields and forests. Physiological Ento-mology, 25(3), 211-222.

Nelson, M. E., & MacIver, M. A. (2006). Sensoryacquisition in active sensing systems. Journalof Comparative Physiology A, 192, 573-586.

Nielsen, C. (2001). Animal evolution: Interrela-tionships of the living phyla (2nd ed.). Oxford:Oxford University Press.

Noe, A (2004). Action in perception. Cambridge,MA: MIT Press.

Northcutt, R. G. (2002). Understanding verte-brate brain evolution. Integrative and Compar-ative Biology, 42(4), 743-756.

Northcutt, R. G., & Gans, C. (1983)' The gene-sis of neural crest and epidermal placodes: Areinterpretation of vertebrate origins. Quar-terly Review ofBiology, 58(1), 1-28.

O'Keefe, J., & Burgess, N. (2005)' Dual phaseand rate coding in hippocampal place cells:Theoretical significance and relationship toentorhinal grid cells. Hippocampus, 15, 853-866.

O'Keefe, 1., & Nadel, 1. (1978). The hippocampusas a cognitive map. New York: Oxford Univer-sity Press.

O'Regan, J. K., & Noe, A. (2001). A sensorimo-tor account ofvision and visual consciousness.Behavioral and Brain Sciences, 24(5),977.

Parsons, S., Thorpe, C. W., & Dawson, S. M.(1997)' Echolocation calls of the long-tailed

bat: A quantitative analysis of types of calls.Journal ofMammalogy, 78(3),

Paulin, M. G. (2005)' Evolutionary origins andprinciples of distributed neural computationfor state estimation and movement control invertebrates. Complexity, 10(3), 56-65.

Pfluger, H. 1., & Menzel, R. (1999). Neuroethol-ogy, its roots and future. Journal of Compara-tive Physiology A, 185(4), 389-392.

Rasnow, B. (1996). The effects of simple objectson the electric field ofApteronotus. Journal ofComparative Physiology A, 178(3), 397-411.

Rasnow, B., & Bower, 1. M. (1996). The electricorgan discharges of the gymnotiform fishes: 1.Apteronotus leptorhynchus. Journal of Compar-ative Physiology A, 178(3), 383-396.

Reed, E. S. (1996). Encountering the world: Towardan ecological psychology. Oxford: Oxford Uni-versity Press.

Reichardt, W., & Poggio, T. (1976). Visual con-trol of orientation behavior in the fly. Part 1.A quantitative analysis. Quarterly Reviews ofBiophysics, 9(3), 311-375.

Reichardt, W., & Wenking, H. (1969). Opti-cal detection and fixation of objects by fixedflying flies. Naturwissenschaften, 56(8), 424-425·

Ross, C. F. (2000). Into the light: The origin ofAnthropoidea:. Annual Review ofAnthropology,29, 147-194.

Rowlands, M. (1999). The body in mind: Under-standing cognitive processes. Cambridge: Cam-bridge University Press.

Russell, 1. 1., & Kossl, M. (1999). Micromechanicalresponses to tones in the auditory fovea ofthe greater mustached bat's cochlea. JournalofNeurophysiology, 82(2), 676-686.

Schnitzler, H. u., Moss, C. F., & Denzinger, A.(2003). From spatial orientation to food acqui-sition in echolocating bats. Trends in Ecology& Evolution, 18(8), 386-394.

Semendeferi, K., Armstrong, E., Schleicher, A.,Zilles, K., & van Hoesen, G. W. (2001). Pre-frontal cortex in humans and apes: A com-parative study of area 10. American Journal ofPhysical Anthropology, 114(3), 224-241.

Sharpee, T. 0., Sugihara, H., Kurgansky, A V.,Rebrik, S. P., Stryker, M. P., & Miller, K.D. (2006). Adaptive filtering enhances infor-mation transmission in visual cortex. Nature,439(7079), 936-942·

Sherry, D. F., Vaccarino, A. L., Buckenham, K.,& Herz, R. S. (1989)' The hippocampal com-plex of food-stOring birds. Brain, Behavior,and Evolution, 34(5), 308-317.

MALCOLM A. MACIVER

Smulders, T V., Sasson, A D., & Devoogd, T 1.(1995)' Seasonal variation in hippocampal vol-ume in a food-storing bird, the black-cappedchickadee. Journal of Neurobiology, 27(1), 15-25·

Snyder, 1. B., Burdick, 1. W., Nelson, M. E.,& MacIver, M. A. (2007)' Omnidirectionalsensory and motor volumes in an electricfish. PLoS Biology, 5(n): e30l doi: 10.13711journal.pbio.o050301.

Srinivasan, M., & Ruina, A (2006). Com-puter optimization of a minimal bipedmodel discovers walking and running. Nature,439(7°72),72-75.

Striedter, G. F. (2005). Principles of brain evolu-tion. Sunderland, MA: Sinauer.

Suga, N. (1990, June). Biosonar and neural com-putation in bats. Scientific American, 60-68.

Szabo, T. (1974). Anatomy of the specialized lat-eralline organs of electroreception. In A Fes-sard (Ed.), Eleetroreceptors and otherspecializedreceptors in lower vertebrates: Handbook of sen-sory physiology (Vol. 3, pp. 13-58). New York:Springer.

Tinbergen, N. (1972). The animal in its world:Explorations of an ethologist, 1932-1972. Lon-don: Allen & Unwin.

U.S. National Traffic Highway Safety Admin-istration. (2004). Driving at night can bedeadly. Retrieved September 19, 2005, fromhttp://ntl.bts.govllib/ooo/zoo/z51/deadly.pdf

van Hateren, J. H., Kern, R., Schwerdtfeger, G.,& Egelhaaf, M. (2005), Function and codingin the blowfly H1 neuron during naturalis-tic optic flow. Journal of Neuroscience, 25(17),4343-4352.

Varela, F. J., Palacios, A G., & Goldsmith, T H.(1993). Color vision ofbirds. In H. P. Zeigler &

H. 1. Bischof (Eds.), Vision, brain, and behav-iorin birds (pp. 77-98). Cambridge, MA: MITPress.

Varju, D. (1975)' Stationary and dynamicresponses during visual edge fixation by walk-ing insects. Nature, 255(55°6), 330-332.

Vendittelli, M., Laumond, J. P., & Nissoux, c.(1999). Obstacle distance for car-like robots.IEEE Transactions on Robotics andAutomation,15(4), 678- 691.

Watanabe, A, & Takeda, K. (1963)' The changeof discharge frequency by AC. stimulus ina weak electric fish. Journal of ExperimentalBiology, 40, 57-66.

Wilson, R. A (1994). Wide computationalism.Mind, 103(411), 351-372.

Wilson, R. A (1995). Cartesian psychology andphysical minds: Individualism and the sciencesofthe mind. Cambridge: Cambridge UniversityPress.

Wotton,1. M., & Simmons, 1. A (2000). Spectralcues and perception of the vertical positionof targets by the big brown bat, Eptesicus !us-cus. Journal ofthe Acoustical Society ofAmerica,107(2), 1034-104l.

Wotton, 1. M., Haresign, T, & Simmons, J.A. (1995)' Spatially dependent acoustic cuesgenerated by the external ear of the bigbrown bat, Eptesicus !uscus. Journal of theAcoustical Society of America, 98(3), 1423-445·

Zhang, 1., & Wang, H. (2005), The effect ofexternal representations on numeric tasks.Quarterly Journal of Experimental Psychology,58A(S), 817-838.

Zupanc, G. K. H. (2004). Behavioral neurobiology:An integrative approach. Oxford: Oxford Uni-versity Press.