Dr Chris Melhuish explains how the natural world has influenced his work in robotics

"I have been studying collective robotic systems for a number of years and was intrigued with the idea of what would happen if one used groups of tiny robots to perform building tasks. The problem with very small robots is that it limits such things as the amount of computation, sensing, communication and power – you have a severely limited (but probably cheap) robot. I was now faced with the task of making a bunch of these simple robots do something smart.

About six years ago I started looking at social insects as a possible source of inspiration. Social insects accomplish impressive tasks (check out a termite mound!) without a central controller; they use a decentralized scalable communication system. In fact I was particularly impressed with the behaviour of Leptothorax ants. These tiny ants, living in small cracks, are able to sort their brood and create defensive walls of sand granules and yet not one ant appears to have a ‘blue print’ for the construction. How do they do that? It seems that they use some kind of ‘emergent control’ strategy – complexity from simplicity! The consequence of each ant carrying out simple behaviours, stimulated by sensing their local environment, creates a structure at the macroscopic level. I was therefore interested to see if we could employ such emergent strategies in a group of simple robots.

I was aware that other scientists were also interested in this problem and one team in particular had made a group of simple robots collect a set of dispersed objects into one pile. They had done so by implementing three simple rules in their robots. I first recreated this mechanism in my own robot group and then started to look in more detail at the sorting problem – how could different classes of objects be sorted by these very simple, limited robots? Trying out new rule sets with robots took a huge amount of time with many experimental runs lasting up to 10 hours or more.

On many occasions we had to abort an experiment to tinker with a rule set and start again. I recall once that after many hours of observation I tried a particular ‘tweak’ to a rule with the result being a partial sorting of two types of object. This looked promising and following many more hours of experimentation the robots were able to sort two types of object using only four simple rules.

This initial success then lead to further success in sorting larger numbers of different types of object. We have now built a simulation of the robot group in order to speed up the experimentation cycle andwe have also recreated our earlier results with the real robots, which gives us some confidence when we try out new ideas on simulated robots. We always go back to the real robot to validate our results – cautioned by a famous roboticist who once warned that ‘simulations are doomed to succeed’!

We are building on this link between biology and robotics. The four strong team now comprises two biologists and two roboticists. The benefits are mutual; biologists want to understand natural mechanisms and roboticists have many questions which invite exploration by biologists.

We are constructing an ‘ant lab’ within our robotics laboratory where we can study Leptothorax ants; recording and analyzing their individual behaviour in an attempt to understand what mechanisms they are employing.

In particular, we will be looking at how the ants sort objects into concentric ring formations. These mechanisms (or those based on these mechanisms) can then be tried out using real and simulated robots.

I tend to be somewhat reluctant to predict applications for the work. Science is riddled with examples of the results of a study in one field being applied in another. Certainly, if we choose to make very small robots then there will be serious constraints implicit in such a system and this will have value – perhaps tiny robots carrying out sensing or surgical tasks within the body?

The field of collective minimalist robots is not restricted to the small scale. It may also provide answers for the control and coordination of larger sized robots. For example, large numbers of cheap (therefore probably relatively simple) robots could be employed in searching or sensing tasks; for example a monitoring pollution in an estuary. We are also currently working on the problem of how to make a group of simple robots with a very limited communication capacity dynamically self-organise into particular structures without the use of a central controller. This work might lead to the development of self-organised shapes for sensing arrays or even aerials."

Chris Melhuish is Director of the Intelligent Autonomous Systems (IAS) Laboratory in the Faculty of Computing, Engineering and Mathematical Sciences at the University of the West of England, Bristol. He has a degree in Geology, an MSc in Computer Science and a PhD in robotics. He is a member of the British Computer Society and is a chartered engineer. The IAS lab comprises a 20 strong team of researchers in fields studying minimalist robotics, aerial robot formation, distributed sensing, inter-robot wireless communication and energy autonomy.

OU Courses

OU Science Faculty website
Practising Science (SXR103)
Discovering Science (S103)

Back to top