Evolving Agent Swarms for Clustering,Patch Sorting, and Annular Sorting

Colonies of social insects are capable of solving complex tasks that far exceed the abilities of each individual insect. The colonies do not use any supervisor or blueprint to organize their work, instead the solutions emerge from the interactions between the insects and their environment. Two of the tasks that social insects perform are clustering of corpses and sorting of brood. In this thesis we describe our work with creating swarms of simple agents that perform similar tasks. Previous research within this field has hand-coded the behavior of the individual agents and then seen if a swarm of the agents is capable of solving the desired task. We take a different approach and evolve the individual behavior by evaluating the patterns that are formed by the swarm. We evolve swarms to solve three different types of tasks: Clustering, patch sorting, and annular sorting. The first two tasks involves the grouping of identical objects, and the grouping of different types of objects into separate groups. Annular sorting involves the creation of a targe-like structure that contains circular bands. This task has not been solved successfully in the past, and we are able to create a dense and well separated structure