Decentralised Autonomic Self-Adaptation in a Foraging Robot Swarm

The deployment of a swarm of robots in domains, such as mine clearance or search and rescue operations, requires that they are self-adaptive in order that they may adjust to unforeseen events and up to date information. Much of the research on swarm self-adaptation focuses on the adaptation of individual swarm behaviour, however a top-down approach may allow a swarm to adjust its behaviour on the basis of the combined knowledge of the swarm. This research looks at producing a decentralised autonomic manager to handle such adaptation in a task of foraging robots, by adjusting the range over which robots broadcast help requests based on the perceived density of the swarm. First, the robots are tasked with recognising the initial situation, before responding to two possible events which alter the scenario, namely the destruction of a proportion of the swarm, and a change in the effective communication range. A centralised system is first developed as an idealised system with full swarm knowledge, and then a decentralised version is created to perform the same role on a per-robot basis, using only the information available to it. The performance of the swarm using each autonomic manager is compared against the performance when using a fixed broadcast range identified to be most suitable for the initial circumstances. It is found that both approaches are capable of recognising the initial situation, and of responding to events, however the effectiveness of the response may depend upon additional parameters not taken into account here. The decentralised autonomic manager presented is also found to require the ability to dynamically alter its own parameters in order to be of use.