Influence of a Social Gradient on a Swarm of Agents Controlled by the BEECLUST Algorithm

Agents controlled by a swarm algorithm interact with each other so that they have collective capabilities that a single agent does not have. The bio-inspired swarm-algorithm “BEECLUST” has the aim to aggregate a swarm at the global optimum even if there are several local optima (of the same type) present. But what about gradients produced of different stimulus types? In this paper, we present the concept of “social stimuli”. We investigate how robots controlled by the BEECLUST-algorithm react to a social stimulus which is created by placing immobilized robots in the environment. It shows that the robots controlled by the BEECLUSTalgorithm are able to react on a social stimulus within an environment with a global and a local optimum. Background & Motivation A swarm intelligent system is based only on a few simple rules which lead to a complex collective behaviour (Beni, 2005). Nature provides many swarm intelligent systems for us, where we can adopt algorithms. In earlier experiments such swarm intelligent behaviour of young honeybees was investigated. The preferred temperature of young honeybees is 36◦C which is also the main temperature in the brood nest of a bee-hive (Grodzicki and Caputa, 2005; Heran, 1952). This temperature is very important for the development of young honeybees and can be interpreted as a natural mechanism to confine them to the brood nest. In previous experiments it was found that a single young honeybee (Apis mellifera) is mostly not able to locate the area with its preferred temperature, whereas a group of young honeybees is able to find the right spot collectively (Kernbach et al., 2009; Schmickl et al., 2008). Figure 1(a) and 1(b) show example-results from such experiments. Similar pictures are shown in (Bodi et al., 2012). From this swarm-intelligent behaviour an algorithm was derived: the BEECLUST algorithm (Kernbach et al., 2009). This algorithm is used in autonomous swarmrobots to find the global optimum out of several local optima. Such optimum could be a light gradient or in our case a temperature gradient. What makes the BEECLUST algorithm special is that it has only a few requirements: The robot needs Not Needed Requirements no explicit communication sensors for distancemeasurement no permanent measurement of the gradient one single non-directional sensor for measuring the gradient