Super Ants

Super Ants

Super AntsMatt deWet & David RobsonSymbiotic CoevolutionPrimary research question:Can heterogeneous teams of evolving agents, who depend upon each other for survival, learn to work together?How to test thatEnvironment needed:Two specialized teams of agents, run by NEATDifferent abilities, different rolesCan only survive by working together

Our EnvironmentAnts!Soldiers & WorkersEnvironmental ThreatsSpidersThese love the taste of worker fleshControlled by a static algorithmStarvationGreat at killing spidersNot so great at gathering food

Our Environment (contd)The worldBounded grid of variable sizeRandomly placed foodRandomly spawned enemiesMovementAll entities move at most one space at a time on the gridMovements all take place simultaneously, so no unit has an advantageThe PlanSensorsSoldiers can see nearby enemies and workersWorkers can see nearby food, enemies, and soldiersDesired behaviorSoldiers learn to keep foraging workers safeHow can we tell?Overall fitness?InspectionThe ExperimentControl Evolve the two groups separately, then stick them together and see how they doExperimentEvolve the two populations together, observe behaviorVariations:Pre-evolved or un-evolved brains.

Current Work

Current WorkCurrent fitness functionsSoldiersfitness: Spiders killedWorkersfitness: How much food is eaten

A video!

Future workShared fitnessAny shared fitness among all agents in one population is nullified, because only relative fitness is used to determine who reproduces.Alternate fitness functionsSlightly more engineeredUpdated sensorsAdd nearby antsBlob sensors

Questions? Ideas?Funny ant stories?