Evolving Swarm Formations for Odour Source Localisation

Abstract

Odour source localisation is a hard problem with many applications. Over the years, researchers have drawn inspiration from Nature to devise many single-robot approaches. Swarm approaches have been growing in popularity, as they offer redundancy to the loss of agents, flexibility, scalability and enable experimenters to employ simpler robots. Many existing swarm approaches make use of robot formations. In this work, we focus on optimising the shape of a swarm formation for finding and tracking odour plumes. We do so by using a genetic algorithm, thus avoiding the cumbersome trial-and-error process that experimenters typically follow to hand-design the formations. The swarm is guided by a leader, which is controlled by a bio-inspired search strategy using the perceptions of the entire swarm. The results show that the evolved formations of three and five robots consistently outperform a single robot and that the best evolved three robot formation is more successful than the hand-designed swarms of three and five robots. As a result, one could opt by using the evolved three robot formation, minimizing the amount of robots needed. Conversely, in case there is a high risk of loss of robots, the evolved five robot formation could be preferable.