Swarm robotics refers to the design and coordination of large numbers of simple physical robots. Uses include environmental applications such as pollution monitoring in the oceans using aquatic robots to monitoring deforestation using drones; conducting search and rescue following earthquakes; activities that demand cheap design due to risk of loss such as mining and most recently in the medical domain using nano-scale robots.
Swarms can be envisaged as complex adaptive systems in which global behavior emerges from local interactions via self-organisation. Programming such behavior is challenging given the large-scale and distributed nature of swarms; in addition, self-organisation also needs to be capable of occurring ‘on the fly’ in response to changes in the environment, e.g. due to failure of swarm members, or changes in the operating environment/user requirements. One approach to achieving the required self-organisation is through a novel theory called Guided Self-Organisation (GSO), which is postulated to be applicable of regulating global (system-wide) behaviour - however, no practical version currently exists. This proposal addresses this issue through developing an algorithmic version of GSO that combines self-supervising and unsupervised learning algorithms, and evaluating its effectiveness in achieving user-goals in the face of change in both simulated and physical swarms