Relative-Value-Based Randomness Algorithm for Swarm Robots Working in Heterogeneous Environments

This paper presents an algorithm, which improves the probability of success for simple swarm robots implementing an extended version of the leaf curling task – one part of the nest building activity of weaver ants [Phan and Russell, 2009, 2010]. The robots are required to have only very basic sensing abilities, and are allowed to transfer information using sematectonic stigmergy – a cue-based indirect communication method. The task of such simple swarm robots is to find and collaborate to manipulate successfully one of the most suitable (easiest) objects in environments where objects have multiple levels of difficulty of some physical property. In earlier work, the original leaf folding task with 2 levels of difficulty of objects was successfully solved using the “exploring rule” [Phan and Russell, 2009]. However, in environments with greater heterogeneity of objects, the exploring rule demonstrated its disadvantages, by achieving very low task completion rates. To address this defect without adding complexity to the hardware of the robots, in this paper we describe in detail an algorithm called relative-value-based randomness (noise). Essentially, this algorithm increases the completion rate of simple swarm robot tasks by encouraging the robots to make a more effective exploration of their complex working environment and to concentrate at a suitable location. The algorithm was developed using physical robots and verified by a series of tests using a simulation model.