A general architecture for robotic swarms

A general architecture for robotic swarms

Swarms are large groups of simplistic individuals that collectively solve disproportionately complex tasks. Individual swarm agents are limited in perception,
mechanically simple, have no global knowledge and are cheap, disposable and fallible. They rely exclusively on local observations and local communications. A swarm has no centralised control.

These features are typifed by eusocial insects such as ants and termites, who construct nests, forage and build complex societies comprised of primitive agents.

This project created the basis of a general swarm architecture for the control of insect-like robots. The Swarm Architecture is inspired by threshold models
of insect behaviour and attempts to capture the salient features of the hive in a closely defined computer program that is hardware agnostic, swarm size indifferent and intended to be applicable to a wide range of swarm tasks.

This was achieved by exploiting the inherent limitations of swarm agents. Individual insects were modelled as a machine capable only of perception, locomotion and manipulation. This approximation reduced behaviour primitives
to a fixed tractable number and abstracted sensor interpretation. Cooperation was achieved through stigmergy and decisions made via a behaviour threshold model.

The Architecture represents an advance on previous robotic swarms in its generality - swarm control software has often been tied to one task and robot configuration. The Architecture's exclusive focus on swarms, sets it apart from
existing general cooperative systems, which are not usually explicitly swarm orientated.

The Architecture was implemented successfully on both simulated and real-world swarms.