SOFTWARE TECHNOLOGIES

homogeneous. Intelligent swarms can also comprise heterogeneous elements Swarms from the outset, reflecting different capabilities as well as a possible social structure. and Swarm Researchers have used agent swarms as a computer modeling tech- nique and as a tool to study complex systems. Simulation examples include Intelligence bird swarms and business, economics, and ecological systems. In swarm sim- Michael G. Hinchey, Loyola College in Maryland ulations, each agent tries to maximize Roy Sterritt, University of Ulster its given parameters. Chris Rouff, Lockheed Martin Advanced Technology Laboratories In terms of bird swarms, each bird tries to find another to fly with, and then flies slightly higher to one side to reduce drag, with the birds eventually forming a flock. Other types of swarm simulations exhibit unlikely emergent behaviors, which are sums of simple Intelligent swarm technologies solve individual behaviors that form com- complex problems that traditional plex and often unexpected behaviors approaches cannot. when aggregated. Swarm intelligence Gerardo Beni and Jing Wang intro- duced the term swarm intelligence in a 1989 article (“Swarm Intelligence,” Proc. 7th Ann. Meeting of the Robotics e are all familiar with working together to achieve a goal Society of Japan, RSJ Press, 1989, pp. swarms in nature. The and produce significant results. 425-428). Swarm intelligence tech- word swarm conjures Swarms may operate on or under the niques (note the difference from intel- up images of large Earth’s surface, under water, or on ligent swarms) are population-based W groups of small insects other planets. stochastic methods used in combina- in which each member performs a torial optimization problems in which simple role, but the action produces SWARMS AND INTELLIGENCE the collective behavior of relatively sim- complex behavior as a whole. The Swarms consist of many simple enti- ple individuals arises from their local emergence of such complex behavior ties that have local interactions, interactions with their environment to extends beyond swarms. Similar com- including interacting with the envi- produce functional global patterns. plex social structures also occur in ronment. The emergence of complex, Swarm intelligence represents a meta- higher-order animals and insects that or macroscopic, behaviors and the heuristic approach to solving a variety don’t swarm: colonies of ants, flocks ability to achieve significant results as of problems. of birds, or packs of wolves. a team result from combining simple, Swarm robotics refers to the appli- These groups behave like swarms in or microscopic, behaviors. cation of swarm intelligence tech- many ways. Wolves, for example, niques to the analysis of activities in accept the alpha male and female as Intelligent swarms which the agents are physical robotic leaders that communicate with the Intelligent swarm technology is devices that can effect changes in their pack via body language and facial based on aggregates of individual environments based on intelligent expressions. The alpha male marks his swarm members that also exhibit inde- decision-making from various input. pack’s territory and excludes wolves pendent intelligence. Members of the The robots can walk, move on that are not members. intelligent swarm can be heteroge- wheels, or operate under water or on Several areas of computer science neous or homogeneous. Due to their other planets. have adopted the idea that swarms differing environments, members can can solve complex problems. For our become a heterogeneous swarm as SWARMS APPLICATIONS purposes, the term swarm refers to a they learn different tasks and develop Practitioners in fields such as tele- large group of simple components different goals, even if they begin as phone switching, network routing,

April 2007 111 SOFTWARE TECHNOLOGIES

pp. 43-53). The models applied artificial neural networks, k-nearest neighbor, and kernel regression techniques. Binary and niching particle swarms solved fea- Asteroid belt ture selection and feature weighting problems. Particle swarms also have influ- enced the computer animation field. Rulers Lagrangian point Rather than scripting the path of each Asteroid(s) habitat individual bird in a flock, Craig W. Workers Messengers Reynolds’ Boids project as described in “Flocks, Herds, and Schools: A Workers Workers Distributed Behavioral Model” (Proc. Earth 14th Ann. Conf. Computer Graphics and Interactive Techniques, ACM Press, X-ray worker 1987, pp. 25-34) elaborates on a par- Messenger ticle swarm using simulated birds as the particles. The simulated flock’s Mag worker aggregate motion behaves much as a real flock would in nature—the dense IR worker interaction comprises the relatively simple behaviors of each of the simu- lated birds choosing its own path.

Figure 1.Prospecting asteroid mission overview.A transport ship launched from Earth Ant colony optimization travels to a point in space where gravitational forces on small objects are all but negligi- Eric Bonabeau, Marco Dorigo, and ble.From the Lagrangian point,1,000 spacecraft will be launched into the asteroid belt, Guy Theraulaz reported much success forming subswarms under the control of a leader or ruler.The subswarms will collect data with their pioneer efforts using social on asteroids of interest,relaying it back to rulers,which ultimately send it back to Earth. behavior patterns of ant colonies to model difficult combinational opti- data categorizing, and shortest-path Eliminating the need for robots to mization problems in Swarm Intelli- optimizations—among others—are have a priori knowledge of the envi- gence: From Natural to Artificial Sys- investigating swarm behavior for ronment or direct communication tems (Oxford Univ. Press, 1999). In potential use in applications. with each other is key to this model. their work, artificial ants travel through a problem graph depositing artificial or BioTracking Particle swarm optimization digital pheromones to enable other ants As part of the BioTracking project at PSO is a global optimization algo- to determine more optimal solutions. the Georgia Institute of Technology, rithm for dealing with problems in Ant colony optimization has solved the researchers have been studying the which a point or surface in an n- traveling salesman problem, which behavior of bees (T. Balch et al., “How dimensional space best represents a investigates the shortest route to several A.I. and Multi-Robot Systems Re- solution. Potential solutions are plot- cities and the subsequent return to a search Will Accelerate Our Under- ted in this space and seeded with an starting point, as well as network and standing of Social Animal Behavior,” initial velocity. Particles move through Internet optimizations. Proc. IEEE, July 2006, pp. 1445- the solution space, and certain fitness 1463). criteria evaluate them. Over time, par- Unmanned underwater vehicles To expedite the understanding of ticles accelerate toward those with University of California, Berkeley, how large-scale robust behavior better fitness values. researchers are studying networks of emerges from the simple behavior of Penn State University researchers have unmanned underwater vehicles. Each individuals, the project videotaped focused on particle swarms for the devel- UUV relies on the same template bees’ behavior over time, using a com- opment of quantitative structure activ- information containing plans, sub- puter vision system to analyze data on ity relationships models used in drug plans, and its own local situation map the insects’ sequential movements to design (W. Cedeño and D.K. Agrafiotis, to make independent decisions. The encode the location of food supplies. “Combining Particle Swarms and k- UUVs, however, cooperate in the net- The intention was to use bees’ behav- Nearest Neighbors for the Development work to conduct, for example, group ior models to improve simple robot of Quantitative Structure-Activity pursuit strategy experiments in a shal- teams capable of complex operations. Relationships,” Biocomputing, 2003, low water pool. They can identify ves-

112 Computer sels of interest and pursue them in missions. The three submissions in the Given that many of the spacecraft environments in which a larger under- autonomous nanotechnology swarm could collide with one another or with water vessel would be destroyed. (ANTS) concept mission deploy mul- asteroids and become lost, multiple- tiple spacecraft to provide backups spacecraft missions offer greater likeli- Swarmcasting and ensure survival in space. hood of survival and flexibility than A technique that exploits the In one incarnation, a Saturn single-spacecraft missions. Addition- acceleration of distributed download- autonomous ring array will launch ally, the self-directed swarm will ing to provide high-resolution video, 1,000 picoclass spacecraft with spe- exhibit intelligence, which is critical audio, and peer-to-peer data streams, cialized instruments—organized as 10 since round-trip delays in communica- swarmcasting also significantly re- subswarms—to perform in situ explo- tion from Earth can stretch upward of duces needed bandwidth. ACTLab’s ration of Saturn’s rings to understand 40 minutes. The mission could be lost Alluvium project at the University of their constitution and formation. before ground control is notified of a Texas at Austin powers ACTLab TV The lander amorphous rover problem. (http://actlab.tv), a concept personal antenna ANTS application is a lunar- TV station, using peer-to-peer media base-activities submission that streaming software. Essentially, it exploits new NASA-developed tech- ature-inspired intelligent swarm applies the swarm analogy to break nologies in the miniaturized robotics technology deals with complex down video files into small pieces so field. Forming the basis for launching N problems that might be impossi- that the system can download com- landers to the moon from remote ble to solve using traditional technolo- ponents from several machines simul- sites, LARA also uses innovative tech- gies and approaches. This field has taneously. Thus, the user can start niques to move rovers in an amoe- spawned a highly successful conference watching the video before the down- boid-like fashion over the moon’s series, the IEEE Swarm Intelligence load completes. uneven terrain. Symposium, the fourth edition of which Swarmcast (www.swarmcast.com), The prospecting asteroid mission takes place this April in Hawaii. This a commercial company, supports involves launching a swarm of technology has even infiltrated pop cul- delivery of large amounts of data autonomous picoclass spacecraft ture and is the subject of Michael over networks using similar concepts (approximately 1 kilogram) to ex- Crichton’s science fiction novel, Prey. ■ and strives to be a significant con- plore the asteroid belt for asteroids tributor to the next generation of with certain characteristics. Figure 1 Michael G. Hinchey is a professor of Internet TV. provides an overview of the PAM computer science at Loyola College in mission concept. In this submission, Maryland. Contact him at mhinchey@ Paintable computers a transport ship launched from Earth loyola.edu. Closely related to the swarm idea is will travel to a point in space where the concept of a paintable computer, gravitational forces on small objects, Roy Sterritt is an academic and as Bill Butera proposed in his PhD dis- such as picoclass spacecraft, are all researcher in the Computer Science sertation (“Programming a Paintable but negligible. Assembled en route Research Institute at the University of Computer,” doctoral dissertation, from Earth, 1,000 spacecraft will be Ulster. Contact him at r.sterritt@ MIT Media Laboratory, MIT, 2002). launched from the Lagrangian point ulster.ac.uk. The idea is that several thousand into the asteroid belt. The space- parasitically powered and pinless inte- craft—equipped with specialized Chris Rouff is manager, Technology and grated circuits—each the size of a instruments—will form subswarms to Business Development, at Lockheed Mar- match head—with an onboard micro- collect relevant data from asteroids of tin Advanced Technology Laboratories. processor, small memory, and wireless interest. Contact him at [email protected]. connectivity can be suspended in a viscous material and then “painted” on various surfaces. Once exposed to a power source, the components or Renew your pfrags self-organize and communicate with external systems via physical IEEE Computer Society contact with objects fitted with a transceiver that defines a suitable com- membership today! munication protocol.

NASA’s use of swarms NASA has been investigating www.ieee.org/renewal swarms for future space exploration

April 2007 113