An approach to time- and space-differentiated pattern formation in multi-robot systems Tim Taylor 1 , Peter Ottery 1 and John Hallam 2,1 1 Institute of Perception, Action and Behaviour, 2 The Maersk Mc-Kinney Moller Institute, School of Informatics, University of Edinburgh, University of Southern Denmark, JCMB, The King’s Buildings, Mayfield Road, Campusvej 55, Edinburgh EH9 3JZ, U.K. DK-5230 Odense M, Denmark tim@tim-taylor.com Abstract We consider the problem of non-trivial pat- tern formation in decentralised multi-robot sys- tems, and, in particular, how to achieve time- and space-varying behaviour. To tackle the problem, we explore the idea of evolving the fine-level regu- lation of an underlying self-organising controller. Results from simulation show the promise of the approach: we demonstrate a robot cluster that can stably maintain two different spatial patterns, switching between the two upon sensing an exter- nal signal; we also demonstrate a cluster in which individual robots develop differentiated states de- spite having identical controllers (which could be used as a starting point for functional specialisa- tion of robots within the cluster). The controller was developed with a particular hardware plat- form in mind—the underwater HYDRON robots developed by the HYDRA consortium (an EU Fifth Framework project). We discuss the im- plementation of the controller on this and other multi-robot platforms comprising free-moving in- dividual robots, and suggest possible simplifica- tions of the design. This work could eventually have applications in various situations that re- quire robust, complex self-organising behaviour in a collection of free-moving robots, e.g. in space, underwater and nano-scale systems. 1. Introduction There has been much interest in decentralised multi- robot systems in recent years, due to their potential advantages in many applications over more traditional, monolithic architectures (Arai et al., 2002). The goal is to design systems that can accomplish their tasks more reliably, faster and/or cheaper than could be achieved by a single more complex robot. The general challenge is to develop controllers for the individual robots such that the group as a whole performs the desired higher-level task through the co-ordinated action of the individuals. In the current work we are specifically interested in multi-robot systems comprising a large number of fairly simple, free-moving robots with limited individual capacity for sensing, actuation and com- munication. The target hardware is described in Section 2. A variety of decentralised controllers for free-moving systems have been proposed in the literature, e.g. (Holland and Melhuish, 1999, Fredslund and Matari´ c, 2002, Nembrini et al., 2002, S ¸ahin et al., 2002, Quinn et al., 2003). In contrast to these previous studies, the present work concentrates on providing an underlying self-organising system for robust pattern formation (a similar approach was suggested by (Spears and Gordon, 1999)). We advance on previous work by extending the complexity of the goal tasks. This is achieved by allowing the behaviour of individual robots to be influenced by communication from neighbouring robots or by detection of signals from the environment. We use a genetic algorithm to evolve controllers to perform particular tasks. There is insufficient space in this paper to provide all the details of the system; the interested reader is referred to (Taylor et al., 2007) for full details. Here we highlight the general approach taken, describe some representa- tive results, and discuss possible simplifications of the approach and potential applications. 2. Robot Hardware and Simulation This work was conducted as part of a project to de- velop distributed controllers for a group of real un- derwater robots called HYDRONs (see (Taylor, 2004, Østergaard et al., 2005)). Each HYDRON unit is a sphere of approximately 11cm diameter, with control of translational movement in three dimensions (provided by a system for impelling/ejecting water for horizon- tal movement, and a buoyancy control system for ver- tical movement). The units also have integrated depth