Hardware variation in robotic swarm and behavioural sorting with swarm chromatography

Abstract

Social insects can achieve remarkable outcomes, various examples can be found in ants, bees, etc. Inspired by social insects, swarm robotic research considers coordinating a group of relatively simple and autonomous robots to finish tasks collaboratively based on direct or indirect interactions. Such systems can offer advantages of robustness, flexibility and scalability, just like social insects.

For many years, various researchers have endeavoured to design intelligent artificial swarms and many hardware-based swarm robots have been implemented. One assumption that made by a majority of swarm robotic researchers, particularly in software simulation is that a robotic swarm is a group of identical robots, there is no difference between any two of them. However, differences among hardware robots are unavoidable, which exist in robotic sensors, actuators, etc. These hardware differences, albeit small, can affect the robots’ response to the environment. Moreover, hardware differences can provoke robots’ heterogeneity which then profoundly influence swarm performance due to the non-linearity in the controller and uncertainty in the environment. Nevertheless, questions about how hardware differences influence swarm performance and how to make use of them remain a research challenge.
In this work, the issue of hardware variation in swarm robots is investigated. Specifically swarm robots with hardware variations are modelled and simulated in a line following scenario. It is found that even small hardware variations can result in behavioural heterogeneity. Although the variations can be compensated by the software controller in training, the hardware variations and resulting differences in training are amplified in the interactions between the robot and the environment.

To know how exactly hardware variation influence robotic behaviours, a novel approach, inspired by the chromatography method in chemistry, is proposed to sort swarm robots according to their hardware circumstances. This method is based on a large number of interactions between robots and the environment. Individual robot’s unique hardware circumstance determines its unique decision making and reaction during each robotic controlling step, and these unique microscopic reactions accumulate and contribute to the robot’s macroscopic behaviour. The behavioural sorting results show that the behaviour of an individual robot is not determined by a single parameter but by the combination of multiple hardware factors. Different combinations of hardware parameters can help robots achieve similar behaviours.

The efficiency of the behavioural sorting method is investigated, particularly the influence of the robot’s controller and environmental factor. By simulating various combinations of robots with different integration lengths of the controller and arenas with different pattern densities, it is discovered that if the robots’ ability to memorise previous events is coupled with the density of the sorting arena, better sorting results can be achieved.

This work is regarded as an initial investigation into the issue of unavoidable hardware differences between swarm robots. Given the research outcome and that real swarms will necessarily show hardware variations, it is therefore necessary to contemplate current swarm algorithms in the context of diverse robot populations. In addition, a new research field of swarm chromatography for sorting robotic behaviours to improve swarm efficiency is initiated.

More information

Identifiers

Catalogue record

Export record