Complexity measures in automatic design of robot swarms: An exploratory study

The design of control software for robot swarms is a challenging endeavour as swarm behaviour is the outcome of the entangled interplay between the dynamics of the individual robots and the interactions among them. Automatic design techniques are a promising alternative to classic ad-hoc design procedures and are especially suited to deal with the inherent complexity of swarm behaviours. In an automatic method, the design problem is cast into an optimisation problem: the solution space comprises instances of control software and an optimisation algorithm is applied to tune the free parameters of the architecture. Recently, some information theory and complexity theory measures have been proposed for the analysis of the behaviour of single autonomous agents; a similar approach may be fruitfully applied also to swarms of robots. In this work, we present a preliminary study on the applicability of complexity measures to robot swarm dynamics. The aim of this investigation is to compare and analyse prominent complexity measures when applied to data collected during the time evolution of a robot swarm, performing a simple stationary task. Although preliminary, the results of this study enable us to state that the complexity measures we used are able to capture relevant features of robot swarm dynamics and to identify typical patterns in swarm behaviour.