Centralised Autonomic Self-Adaptation in a Foraging Robot Swarm

The use of robotic swarms in domains, such as space exploration, or search and rescue missions, requires that the swarms be self-adaptive in order to adjust to newly acquired data, and react to unforeseen events. Research on swarm self-adaptation tends to focus on the adaptation of individual agents, however taking a top-down approach can allow for the use of knowledge that is only apparent considering the swarm as a whole. This research makes use of a centralised Autonomic Manager to modify the behaviour of a simulated swarm of foraging robots by adjusting the range over which individual robots broadcast help requests. The swarm is able to learn its own size and the size of the test area, and use that information to guide its decision making, showing the potential for a future decentralised approach. First, the swarm is tasked with recognising the initial situation. Secondly, the swarm must respond to two events which alter the scenario parameters, namely the destruction of a proportion of the swarm, and a change in effective communication range. Performance of the swarm using an Autonomic Manager is compared against that using a fixed broadcast range suited to the initial circumstances. The results show that the swarm can recognise the initial situation and select a suitable broadcast range. It is also capable of recognising the events that occur, but the effectiveness of its response depends on additional parameters in the simulation.