how social insects provide solutions to complex problems - NiSIS

Swarm intelligent systems: how social insectsprovide solutions to complex problemsGuy TheraulazCentre de Recherches sur la Cognition AnimaleCNRS, UMR 5169, Toulouse, FranceNISIS Annual SymposiumTenerife, November 29 - December 1 st , 2006

Large-scale structuresNestEciton burchelli

Large-scale structures

Complex structuresApicotermes lamani

Complex structuresApicotermes lamani

Anthropomorphic hypothesis! The complexity of behaviors andpatterns observed at the colonylevel should be a directconsequence of the individualsability:" to centralize informationabout the environmentalconditions" to build an internalrepresentation of theseconditions and then …" to choose the appropriateactions to perform

A hierarchical and centralized organization ?! Informationcoming from thecolony isgathered andmonitored by thequeen which thendirects workersactivities

The society has no supervisor~ 10 000 - 100 000 neurons(10 billions in man)! Individual insects do not have a mental blueprint of the architecturesthey build or a global representation of the state of the colony! Their cognitive system is not enough powerful for a single individualto assess a global situation, centralize all the information comingfrom its colony and then control the tasks to be done by the otherworkers

Social insects colonies aredistributed information processing systems! Local informationThe rules specifying the interactionsamong insects are executed on the basisof purely local information, without anyknowledge of the global pattern! A limited set of instructionsEach insect is following a small set ofsimple behavioral rules (20 elementarybehaviors in ants)

Social insects colonies aredistributed information processing systems! Local informationThe rules specifying the interactionsamong insects are executed on the basisof purely local information, without anyknowledge of the global pattern! A limited set of instructionsEach insect is following a small set ofsimple behavioral rules (20 elementarybehaviors in ants)! Emergent cooperationComplex collective behaviors emergefrom interactions among individuals thatexhibit simple behaviors. Social insectscan solve problems in a very flexible androbust way

Pierre-Paul Grassé (1895-1985)Stigmergic interactions1959La reconstruction du nid et les coordinationsinter-individuelles chez Bellicositermesnatalensis et Cubitermes sp. La théorie de lastigmergie : essai d’interprétation ducomportement des termites constructeurs.Insectes Sociaux, 6, 41-81“An insect does not control his ownwork. But its ongoing activity is guidedby the by-product of its work”

Stigmergy: invisible writing! Stigmergy occurs when insect’s actions are determined or influenced bythe consequences of another insect’s previous action! This is a form of indirect communication that makes possible thecoordination and regulation of insects activitiesResponseR 4 R 5S 5R 1S 2R 2S 3R 3S 4Stimulus S 1Stoptime! This process leads to an (almost) perfect coordination of the collectivework and gives us the impression that a colony as a whole is followinga pre-defined plan

Nest building in social waspsA stigmergic behavior! Wasp nests are built with wood pulpand plant fibers! Colored blotting paper used asbuilding material makes it possible thevisualization of successive buildingsteps! Individual construction behavior can bestudied in great details such as thewasps decisions to build a new cell onparticular locations on the combNest constructionin Polistinae wasps

The control of nest building in waspsPotential building sites on a comb! The nest structure controls the organization of building activities! To decide where to build a new cell, wasps make use of the informationprovided by the local arrangement of cells on the comb

Construction rules in Polistes waspsProbability to build a new cell on the comb1.00.80.6! New cells are not added randomly tothe existing structure! Wasps have a greater probability toadd new cells to a corner area (3 or 4adjacent walls) than to initiate a newrow by adding a cell on the side of anexisting row (2 adjacent walls)0.40.20.01 2 3 4Number of adjacent walls

Modeling nest buildingBehavior of virtual waspsLocal neighborhoodof the virtual wasp! Wasps are modeled by asynchronousautomata with a stimulus-responsebehavior! Virtual wasps move randomly in a 3-Ddiscrete hexagonal lattice! Virtual wasps only have a localperception of their environment (thefirst 26 neighboring cells close the celloccupied by the wasp)! … and do not have anyrepresentation of the globalarchitecture they build(Theraulaz, G. & Bonabeau, E., Science, 1995)

Modeling nest buildingLocal construction rules used by virtual waspsStimulusResponseStimulusResponse1.000.05! Someconfigurationsof cells triggerthe constructionof a new cell! Constructionrules areprobabilistic1.001.001.000.501.00

Simulations of collective buildingwith a 3D lattice swarmNest architectures obtained by simulationPolistes dominulus

Exploration of the morphospaceNest architectures obtained by simulationAgelaia testacea

Exploration of the morphospaceNest architectures obtained by simulationParachartergus fraternus

Exploration of the morphospaceNest architectures obtained by simulationVespa crabro

Exploration of the morphospaceNest architectures obtained by simulationChartergus chartarius

Exploration of the morphospaceNest architectures obtained by simulationChartergus chartarius

Trail recruitment in antsA self-organized behavior based on stigmergic interactions

Ants looking for food: mass recruitment

Formation of foraging trailsNestFoodsource! As the ants return from the food source to the nest,they lay down a trail of pheromones that can befollowed other ants! Recruited ants lay down their own pheromone onthe trail as well, reinforcing the pathway! Trail formation results from a positive feed-back! Negative feed-back results from the evaporation ofpheromone or the exhaustion of the food source! Trail’s recruitment system enables efficientdecision making

Collective decisionsPath selection toward a food sourceChoice occurs randomly(Deneubourg, J.L. et al., J. Ins. Behav., 1990)

Collective decisionsPath selection toward a food sourceBranch ABranch B

Collective decisionsPath selection toward a food sourceBranch ABranch B

Collective decisionsPath selection toward a food sourceBranch ABranch B

Collective decisionsPath selection toward a food sourceBranch ABranch B

Collective decisionsPath selection toward a food source! Collective decisions in socialinsects arise through thecompetition among differenttypes of information! The trail that succeeds to growfaster will be selected, leadingall the traffic to take place onone of the competing branches! Environmental constraintsinterplay with the positivefeedback

Collective decisionsSelection of the shortest route toward a food source(Goss, S. et al., Naturwissenchaften, 1989)

Collective decisionsSelection of the shortest route toward a food source! Ants first use both paths in equalnumbers, laying downpheromones as they move! Ants taking the shorter pathreturn to the nest faster! The shorter path will then bedoubly marked with pheromone,and will thus be more attractive! Geometrical constraints play akey role in the collectivedecision-making processes thatemerge at the colony level

Self-organization: a key conceptto understand collective behaviorsTrail networksformationDivisionof laborColonythermoregulationNestconstructionSynchronizationof behaviorsComb patternsformation! Self-organization isa set of dynamicalmechanismswhereby structuresor decisions emergeat the global level ofa system frominteractions amongits lower-levelcomponents, withoutbeing explicitlycoded at theindividual level

The ingredients of self-organization! Positive feedback (amplification): theyare simple behavioral 'rules of thumb'that promote the creation of structures! Negative feedback: counterbalancespositive feedback and helps to stabilizethe collective pattern: it may take theform of saturation, exhaustion orcompetition! Amplification of fluctuations: fluctuationsact as seeds from which structuresnucleate and grow. Randomnessenables the discovery of new solutions(Bonabeau, E. et al., TREE, 1997)

The ingredients of self-organization! Positive feedback (amplification): theyare simple behavioral 'rules of thumb'that promote the creation of structures! Negative feedback: counterbalancespositive feedback and helps to stabilizethe collective pattern: it may take theform of saturation, exhaustion orcompetition! Amplification of fluctuations: fluctuationsact as seeds from which structuresnucleate and grow. Randomnessenables the discovery of new solutions! Multiple interactions: enable thestochastic nature of the underlyingmechanisms to produce the appearanceof large and enduring structuresCorpse aggregation in ants(Bonabeau, E. et al., TREE, 1997)

Emergence and optimization of division of laborPolistes dominulus

Distribution of activity levelsas a function of colony size10InactivityWorkActivity levelSize of the colony

Distribution of activity levelsas a function of colony size30The amount of task to be doneis proportional to colony sizeActivity levelHard workersSize of the colony

Distribution of activity levelsas a function of colony sizeBifurcation300Activity levelSize of the colony

Self-organized division of laborResponse threshold reinforcementPositive feed-backYESTask-associatedstimuliResponse thresholdTheraulaz et al., Proc. Roy. Soc. B., (1998)Executetask ?! The more a wasp performs a task, thelower its response threshold withrespect to stimuli associated with thistask! The wasp is more responsive to smallvariation of the stimuli and becomesspecialized in the execution of thecorresponding task

Self-organized division of laborResponse threshold reinforcementOUI YESTask-associatedstimuliExecutetask?! Not performing the task induces anincrease of the response threshold! The wasp will be less responsive tothe stimuli and its probability toperform the task will be lowerNONegative feed-back

Self-organized division of labor! Several insects are incompetition to perform thetaskTask-associatedstimuli! The total amount of workload is proportional to thesize of the colony! Division of labor at thecolony level results fromIndividual learning andcompetition amongindividuals to perform tasks

Self-organized division of laborResponse thresholdsTask-associatedstimuliInactiveindividualsHard-workerstime

Self-organized division of laborActivity level10wasps30wasps300wasps! In a small colony, the amount ofwork to be done is small andlarge fluctuations of work loadoccur with time: wasps have notenough time and no opportunityto undergo a differentiation! As colony size is increasing, theabsolute value of the amount ofwork to be done increases,fluctuations of task associatedstimuli become weaker andgreater are the chance for someindividuals to become hardworkersGautrais et al., J. theor. Biol., (2002)

Sequence of tasksinvolved in nest constructionWaterforagingPolistes dominulusAs colony size increases,individuals become specialized inperforming tasksBuildingon the nestPulpforaging

Influence of colony sizeon individual specializationNumber of wasps in the colony: 30WaterforagingProbability to keepdoing the same taskProbability to switchto another taskWater sharingBuildingon the nestPulpforaging(Karsai, I. & Wenzel, J., PNAS, 1998)

Influence of colony sizeon individual specializationNumber of wasps in the colony : 50WaterforagingBuildingon the nestPulpforaging(Karsai, I. & Wenzel, J., PNAS, 1998)

Influence of colony sizeon individual specializationNumber of wasps in the colony : 300WaterforagingBuildingon the nestPulpforaging(Karsai, I. & Wenzel, J., PNAS, 1998)

Self-organized optimizationof division of labor! The organization of division of labor isadapted to the size of the colony! In a large colony, with high work load,specialized, highly skilled workers increasethe efficiency of the division of labor! Its is better for a small colony to keepgeneralist workers! Self-organization allows for the optimizationof the division of labor

Bio-mimetic models of swarm intelligence! Social insect colonies are decentralizedproblem-solving systems! Many of the daily problems solved by acolony have counterparts in engineeringand computer science! Models of collective behaviors in socialinsects provide us with powerful tools totransfer knowledge about social insects tothe field of intelligent system designroblems.(Bonabeau & Theraulaz, Sci. Am., 2000)

Bio-mimetic models of swarm intelligenceDistributed control in swarm-based robotics! A group of robots may be able toperform tasks without explicitrepresentations (environment, otherrobots …)! A group of simple robots may bemore efficient, much more reliablethan a single complex robot! Using a group of simple robots bringsnew problems (lack of globalknowledge, poor communicationabilities)! Swarm-based robotics relies onstigmergic communication(environment is used ascommunication channel)

Bio-mimetic models of swarm intelligenceStigmergic communication in swarm roboticsCollective level22 mmBehavioralmodelIndividual levelInfrared sensors

Bio-mimetic models of swarm intelligenceStigmergic communication in swarm roboticsLight sensors

Bio-mimetic models of swarm intelligenceStgmergic communication in swarm-based roboticsFoodsourceANest1Probability to chose branch AB0,81robot0,60,40,200 10 20 30 40 50Time (mn)

Bio-mimetic models of swarm intelligenceStgmergic communication in swarm-based roboticsFoodsourceANest1Probability to chose branch AB0,85robots0,60,40,200 10 20 30 40 50Time (mn)

Toward scalable swarm intelligent systems! The binary choice experiment: atest-bed situation for designing newcontrol algorithms! A challenging problem: finding acontrol algorithm for the selfadaptationof the group behavior inresponse to changingenvironmental conditions! What should be the controlalgorithm that would enable agroup of robots to make a stablechoice whatever the size of thegroup?

Toward scalable swarm intelligent systems! The binary choice experiment: atest-bed situation for designing newcontrol algorithms! A challenging problem: finding acontrol algorithm for the selfadaptationof the group behavior inresponse to changingenvironmental conditions! What should be the controlalgorithm that would enable agroup of robots to make a stablechoice whatever the size of thegroup?! Ants got the solution! They use therate of contacts to regulate thetraffic organization

Conclusions and perspectives! Complex colony-level structuresand swarm intelligence of socialinsects emerge from decentralizedinteractions among individuals! Swarm intelligent systems areflexible and robust! Social insects are a rich source ofinspiration to design adaptivedecentralized artificial systems

Acknowledgements# Centre de Recherche sur laCognition Animale, CNRS,Toulouse, FranceRichard BonVincent FourcassiéSimon GarnierJacques GautraisAnne GrimalRaphaël JeansonChristian JostAndrea PernaKarla Vittori# Laboratoire d’Energétique,UPS, Toulouse, FranceStéphane BlancoRichard Fournier# Institut de Recherche enInformatique de Nantes, FrancePascale Kuntz# School of Biological Sciences,The University of Sydney, AustraliaJérôme Buhl# Service d’Ecologie SocialeULB, Bruxelles, BelgiumJean-Louis Deneubourg# Complex Systems Lab,Universitat Pompeu Fabra,Barcelona, SpainRicard SoléSergi Valverde# Autonomous Systems Lab,E.P.F.L., Lausanne, SwitzerlandMasoud AsadpourGilles CaprariFabien Tâche

To learn more about self-organizationand swarm intelligence in social insects(Camazine et al., 2001)Princeton University Press(Bonabeau, Dorigo & Theraulaz, 1999)Oxford University Press