Swarm controlled emergence for ant clustering

Swarm controlled emergence is proposed as an approach to control emergent effects in (artificial) swarms. The method involves the introduction of specific control agents into the swarm systems. Control agents behave similar to the normal agents and do not directly influence the behavior of the normal agents. The specific design of the control agents depends on the particular swarm system considered. The aim of this paper is to apply the method to ant clustering. Ant clustering, as an emergent effect, can be observed in nature and has inspired the design of several technical systems, e.g. moving robots, and clustering algorithms.

Different types of control agents for that ant clustering model are designed by introducing slight changes to the behavioural rules of the normal agents. The clustering behaviour of the resulting swarms is investigated by extensive simulation studies.

It is shown that complex behavior can emerge in systems with two types of agents (normal agents and control agents). For a particular behavior of the control agents, an interesting swarm size dependent effect was found. The behaviour prevents clustering when the number of control agents is large, but leads to stronger clustering when the number of control agents is relatively small.

Although swarm controlled emergence is a general approach, in the experiments of this paper the authors concentrate mainly on ant clustering. It remains for future research to investigate the application of the method in other swarm systems. Swarm controlled emergence might be applied to control emergent effects in computing systems that consist of many autonomous components which make decentralized decisions based on local information.

The particular finding, that certain behaviours of control agents can lead to stronger clustering, can help to design improved clustering algorithms by using heterogeneous swarms of agents.

In general, the control of (unwanted) emergent effects in artificial systems is an important problem. However, to date not much research has been done on this topic. This paper proposes a new approach and opens a different research direction towards future control principles for self‐organized systems that consist of a large number of autonomous components.