April 2007 111

S O F T W A R E T E C H N O L O G I E S

Intelligent swarm technologies solve

complex problems that traditional

approaches cannot.

W e are all familiar withswarms in nature. Theword swarm conjuresup images of largegroups of small insects

in which each member performs asimple role, but the action producescomplex behavior as a whole. Theemergence of such complex behaviorextends beyond swarms. Similar com-plex social structures also occur inhigher-order animals and insects thatdon’t swarm: colonies of ants, flocksof birds, or packs of wolves.

These groups behave like swarms inmany ways. Wolves, for example,accept the alpha male and female asleaders that communicate with thepack via body language and facialexpressions. The alpha male marks hispack’s territory and excludes wolvesthat are not members.

Several areas of computer sciencehave adopted the idea that swarmscan solve complex problems. For ourpurposes, the term swarm refers to alarge group of simple components

working together to achieve a goaland produce significant results.Swarms may operate on or under theEarth’s surface, under water, or onother planets.

SWARMS AND INTELLIGENCESwarms consist of many simple enti-

ties that have local interactions,including interacting with the envi-ronment. The emergence of complex,or macroscopic, behaviors and theability to achieve significant results asa team result from combining simple,or microscopic, behaviors.

Intelligent swarmsIntelligent swarm technology is

based on aggregates of individualswarm members that also exhibit inde-pendent intelligence. Members of theintelligent swarm can be heteroge-neous or homogeneous. Due to theirdiffering environments, members canbecome a heterogeneous swarm asthey learn different tasks and developdifferent goals, even if they begin as

homogeneous. Intelligent swarms canalso comprise heterogeneous elementsfrom the outset, reflecting differentcapabilities as well as a possible socialstructure.

Researchers have used agentswarms as a computer modeling tech-nique and as a tool to study complexsystems. Simulation examples includebird swarms and business, economics,and ecological systems. In swarm sim-ulations, each agent tries to maximizeits given parameters.

In terms of bird swarms, each birdtries to find another to fly with, andthen flies slightly higher to one side toreduce drag, with the birds eventuallyforming a flock. Other types of swarmsimulations exhibit unlikely emergentbehaviors, which are sums of simpleindividual behaviors that form com-plex and often unexpected behaviorswhen aggregated.

Swarm intelligenceGerardo Beni and Jing Wang intro-

duced the term swarm intelligence in a 1989 article (“Swarm Intelligence,”Proc. 7th Ann. Meeting of the RoboticsSociety of Japan, RSJ Press, 1989, pp.425-428). Swarm intelligence tech-niques (note the difference from intel-ligent swarms) are population-basedstochastic methods used in combina-torial optimization problems in whichthe collective behavior of relatively sim-ple individuals arises from their localinteractions with their environment toproduce functional global patterns.Swarm intelligence represents a meta-heuristic approach to solving a varietyof problems.

Swarm robotics refers to the appli-cation of swarm intelligence tech-niques to the analysis of activities inwhich the agents are physical roboticdevices that can effect changes in theirenvironments based on intelligentdecision-making from various input.The robots can walk, move onwheels, or operate under water or onother planets.

SWARMS APPLICATIONSPractitioners in fields such as tele-

phone switching, network routing,

Swarms and SwarmIntelligenceMichael G. Hinchey, Loyola College in Maryland Roy Sterritt, University of UlsterChris Rouff, Lockheed Martin Advanced Technology Laboratories

112 Computer

S O F T W A R E T E C H N O L O G I E S

data categorizing, and shortest-pathoptimizations—among others—areinvestigating swarm behavior forpotential use in applications.

BioTrackingAs part of the BioTracking project at

the Georgia Institute of Technology,researchers have been studying thebehavior of bees (T. Balch et al., “HowA.I. and Multi-Robot Systems Re-search Will Accelerate Our Under-standing of Social Animal Behavior,”Proc. IEEE, July 2006, pp. 1445-1463).

To expedite the understanding ofhow large-scale robust behavioremerges from the simple behavior ofindividuals, the project videotapedbees’ behavior over time, using a com-puter vision system to analyze data onthe insects’ sequential movements toencode the location of food supplies.The intention was to use bees’ behav-ior models to improve simple robotteams capable of complex operations.

Eliminating the need for robots tohave a priori knowledge of the envi-ronment or direct communicationwith each other is key to this model.

Particle swarm optimizationPSO is a global optimization algo-

rithm for dealing with problems inwhich a point or surface in an n-dimensional space best represents asolution. Potential solutions are plot-ted in this space and seeded with aninitial velocity. Particles move throughthe solution space, and certain fitnesscriteria evaluate them. Over time, par-ticles accelerate toward those withbetter fitness values.

Penn State University researchers havefocused on particle swarms for the devel-opment of quantitative structure activ-ity relationships models used in drugdesign (W. Cedeño and D.K. Agrafiotis,“Combining Particle Swarms and k-Nearest Neighbors for the Developmentof Quantitative Structure-ActivityRelationships,” Biocomputing, 2003,

pp. 43-53). The models applied artificialneural networks, k-nearest neighbor,and kernel regression techniques. Binaryand niching particle swarms solved fea-ture selection and feature weightingproblems.

Particle swarms also have influ-enced the computer animation field.Rather than scripting the path of eachindividual bird in a flock, Craig W.Reynolds’ Boids project as describedin “Flocks, Herds, and Schools: ADistributed Behavioral Model” (Proc.14th Ann. Conf. Computer Graphics andInteractive Techniques, ACM Press,1987, pp. 25-34) elaborates on a par-ticle swarm using simulated birds asthe particles. The simulated flock’saggregate motion behaves much as areal flock would in nature—the denseinteraction comprises the relativelysimple behaviors of each of the simu-lated birds choosing its own path.

Ant colony optimizationEric Bonabeau, Marco Dorigo, and

Guy Theraulaz reported much successwith their pioneer efforts using socialbehavior patterns of ant colonies tomodel difficult combinational opti-mization problems in Swarm Intelli-gence: From Natural to Artificial Sys-tems (Oxford Univ. Press, 1999). Intheir work, artificial ants travel througha problem graph depositing artificial ordigital pheromones to enable other antsto determine more optimal solutions.Ant colony optimization has solved thetraveling salesman problem, whichinvestigates the shortest route to severalcities and the subsequent return to astarting point, as well as network andInternet optimizations.

Unmanned underwater vehiclesUniversity of California, Berkeley,

researchers are studying networks ofunmanned underwater vehicles. EachUUV relies on the same templateinformation containing plans, sub-plans, and its own local situation mapto make independent decisions. TheUUVs, however, cooperate in the net-work to conduct, for example, grouppursuit strategy experiments in a shal-low water pool. They can identify ves-

Lagrangian pointhabitat

Earth

Asteroid belt

Asteroid(s)Rulers

Workers Messengers

Workers Workers

MessengerX-ray worker

Mag worker

IRworker

Figure 1. Prospecting asteroid mission overview. A transport ship launched from Earth

travels to a point in space where gravitational forces on small objects are all but negligi-

ble. From the Lagrangian point, 1,000 spacecraft will be launched into the asteroid belt,

forming subswarms under the control of a leader or ruler.The subswarms will collect data

on asteroids of interest, relaying it back to rulers, which ultimately send it back to Earth.

sels of interest and pursue them inenvironments in which a larger under-water vessel would be destroyed.

SwarmcastingA technique that exploits the

acceleration of distributed download-ing to provide high-resolution video,audio, and peer-to-peer data streams,swarmcasting also significantly re-duces needed bandwidth. ACTLab’sAlluvium project at the University ofTexas at Austin powers ACTLab TV(http://actlab.tv), a concept personalTV station, using peer-to-peer mediastreaming software. Essentially, itapplies the swarm analogy to breakdown video files into small pieces sothat the system can download com-ponents from several machines simul-taneously. Thus, the user can startwatching the video before the down-load completes.

Swarmcast (www.swarmcast.com),a commercial company, supportsdelivery of large amounts of dataover networks using similar conceptsand strives to be a significant con-tributor to the next generation ofInternet TV.

Paintable computersClosely related to the swarm idea is

the concept of a paintable computer,as Bill Butera proposed in his PhD dis-sertation (“Programming a PaintableComputer,” doctoral dissertation,MIT Media Laboratory, MIT, 2002).

The idea is that several thousandparasitically powered and pinless inte-grated circuits—each the size of amatch head—with an onboard micro-processor, small memory, and wirelessconnectivity can be suspended in aviscous material and then “painted”on various surfaces. Once exposed toa power source, the components orpfrags self-organize and communicatewith external systems via physicalcontact with objects fitted with atransceiver that defines a suitable com-munication protocol.

NASA’s use of swarmsNASA has been investigating

swarms for future space exploration

missions. The three submissions in theautonomous nanotechnology swarm(ANTS) concept mission deploy mul-tiple spacecraft to provide backupsand ensure survival in space.

In one incarnation, a Saturnautonomous ring array will launch1,000 picoclass spacecraft with spe-cialized instruments—organized as 10subswarms—to perform in situ explo-ration of Saturn’s rings to understandtheir constitution and formation.

The lander amorphous roverantenna ANTS application is a lunar-base-activities submission thatexploits new NASA-developed tech-nologies in the miniaturized roboticsfield. Forming the basis for launchinglanders to the moon from remotesites, LARA also uses innovative tech-niques to move rovers in an amoe-boid-like fashion over the moon’suneven terrain.

The prospecting asteroid missioninvolves launching a swarm ofautonomous picoclass spacecraft(approximately 1 kilogram) to ex-plore the asteroid belt for asteroidswith certain characteristics. Figure 1provides an overview of the PAMmission concept. In this submission,a transport ship launched from Earthwill travel to a point in space wheregravitational forces on small objects,such as picoclass spacecraft, are allbut negligible. Assembled en routefrom Earth, 1,000 spacecraft will belaunched from the Lagrangian pointinto the asteroid belt. The space-craft—equipped with specializedinstruments—will form subswarms tocollect relevant data from asteroids ofinterest.

Given that many of the spacecraftcould collide with one another or withasteroids and become lost, multiple-spacecraft missions offer greater likeli-hood of survival and flexibility thansingle-spacecraft missions. Addition-ally, the self-directed swarm willexhibit intelligence, which is criticalsince round-trip delays in communica-tion from Earth can stretch upward of40 minutes. The mission could be lostbefore ground control is notified of aproblem.

N ature-inspired intelligent swarmtechnology deals with complexproblems that might be impossi-

ble to solve using traditional technolo-gies and approaches. This field hasspawned a highly successful conferenceseries, the IEEE Swarm IntelligenceSymposium, the fourth edition of whichtakes place this April in Hawaii. Thistechnology has even infiltrated pop cul-ture and is the subject of MichaelCrichton’s science fiction novel, Prey. ■

Michael G. Hinchey is a professor ofcomputer science at Loyola College inMaryland. Contact him at mhinchey@loyola.edu.

Roy Sterritt is an academic andresearcher in the Computer ScienceResearch Institute at the University ofUlster. Contact him at r.sterritt@ulster.ac.uk.

Chris Rouff is manager, Technology andBusiness Development, at Lockheed Mar-tin Advanced Technology Laboratories.Contact him at crouff@atl.lmco.com.

April 2007 113

Renew your IEEE Computer Society

membership today!

www.ieee.org/renewal