|dc.description.abstract||This report explores the possibilities of using repellent pheromones in order to achieve efficient dispersion of a swarm of robots. The goal is to have a set of autonomous robots cover an unknown environment by collectively visiting each area once.
Unlike previous research in collective exploration this report aims at achieving intelligent dispersion by relying solely on local perception of repellent pheromones, without any centralized control mechanisms.
Using local sensory input and decentralized controllers promotes better scalability of the swarm, and is well suited for application on physical robots.
A small number of homogeneous autonomous robots are tasked to explore randomly generated environments. The robots possess no previous knowledge of the environments nor the precise whereabouts of other robots. Repellent virtual pheromones serve as a collective memory of previously visited areas, aimed at preventing several robots from searching the same locations twice, thus increasing the exploration efficiency.
The usefulness of repellent pheromones in exploration is addressed through agent-based simulations, with respect to population density, environmental layout, and pheromone evaporation rate.
Experimental results show that repellent pheromones can assist robots in exploring unknown environments more efficiently and consistently. The benefits of using pheromones is shown to be greatest in low population swarms and when environments are large or contain few obstacles. Moreover, pheromones evaporating at a slow rate is shown to provide consistent improvements over rapidly evaporating pheromones.||