Complex instincts

Modelled on the behaviour of rats, robots developed at Sheffield University are designed to tackle relatively complex operations.

Not so long ago robots were considered the stuff of science fiction. But now robots are well into the realm of science fact.

Robotics is a growing industry, and United Nations estimates suggest that by the end of 2007 there will be more than four million robots in use around the world. Robots already play a vital role in defence and security, space exploration and on the production line. They are also becoming increasingly important for entertainment applications and as human companions. But their usefulness doesn’t end there.

According to Dr. Tony Prescott of the Department of Psychology at SheffieldUniversity, robots can also play an important role in the search for answers to one of the most fundamental mysteries of life: The workings of the vertebrate brain.

‘Robots are a kind of physical model,’ explains Dr. Prescott. ‘We are simulating and building robots as a tool to gain a better understanding of what the brain is doing and how it is operating.’

The robots Dr. Prescott uses are a cut above the ordinary. Most robots today are designed to carry out just a single repetitive task, such as tightening bolts on a production line or mowing a lawn. Dr. Prescott’s robots, on the other hand, are built to operate in a much more ‘life-like’ way. Modelled on the behaviour of rats, his robots are designed to tackle relatively complex operations that involve switching between tasks. One, for example, collects cylinders and carries them about. Sounds simple, but in fact a number of tasks are involved; the robot needs to know when to look for cylinders, when to move towards them, and how to locate walls or corners where it should stay and hide.

Complex tasks

By working to develop a robust control system for these multitasking robots, Dr. Prescott and his Sheffield-based colleagues – neuroscientist Professor Peter Redgrave, computational modellers Dr. Kevin Gurney and Dr Mark Humphries – make up one of the few groups internationally whose activities straddle neurobiological research, computational modelling and robot modelling.

Working together they are gaining insights into how vertebrate brains handle everyday, yet relatively complex, tasks. Examples include foraging, where animals must oscillate between several tasks, deciding when and where to seek food and drink and when and where to eat and drink.

‘In these situations there is a trade-off between the various things you could be doing, and there are good reasons why you might be doing any of them,’ Dr Prescott explains. ‘At any one time you are trying to decide which of the alternatives is the most important and to prioritise it.’ This switching behaviour, or action selection, is thought to be carried out in the basal ganglia and the reticular formation, areas in the lower part of the brain.

Because these brain regions are present in all vertebrates, studies of their function in laboratory animals can also provide insights into the role they play in humans and other mammals.

The group are also drawing inspiration for their work from the pioneering studies of artificial neural networks carried out in 1969 by Warren McCulloch at the Massachusetts Institute of Technology, a neuroscientist who was also interested in reverse engineering the brain to understand how it operates. ‘

When we took McCulloch’s model, implemented it and tried to control a robot with it we found it didn’t work very well!’ reports Dr. Prescott. But after working to develop the model further, the group came up with some interesting variations. These turned out to be very good action selectors for the particular tasks that they have been evolved for, and are shedding light on the evolution and role of the reticular formation. ‘

Newly born rats don’t have a functioning basal ganglia, so the instinctive behaviours of rat pups are probably organised in the reticular formation,’ explains Dr. Prescott.

‘Our variations on McCulloch’s model suggest that the reticular formation has evolved the right connectivity in order to be able to allow newborns to express the right behaviours at the right time.’

Additional insights

Other aspects of the group’s work are revealing insights into the workings of the basal ganglia, a part of the brain that comes ‘on line’ after the reticular formation. ‘Our ultimate goal is to look at how the basal ganglia and reticular formation operate together,’ says Dr. Prescott.

‘To some extent what we are trying to do is to understand the brain’s real-time operating system. We want to know how it is deciding which actions to select at any one time and how it is partitioning up the problem of behaviour and control.’

This information has many potential uses.

Because diseases such as Parkinson’s disease, Tourette’s syndrome, Huntington’s chorea, schizophrenia, obsessive-compulsive disorder and attention deficit hyperactivity disorder (ADHD) in children are linked to malfunctions in the basal ganglia, the group’s work will also lend insights into how these diseases work. These, in turn, may help in the development of treatments.

They could also lead eventually to the development of more human-like robots, able to carry out several tasks and switch between them flexibly.

‘We think that if you can understand the rat brain first, then you can make a lot of progress towards understanding the human brain,’ notes Dr. Prescott. ‘Ultimately we’d like to understand the human brain and develop human-like robots that people will feel comfortable interacting with in human environments. But that lies in the future. Our more immediate goals are to understand how the vertebrate brain works and to build rat-like robots.’

Rat meets ratbot: By creating robots that learn much like newborn rats EPSRC researchers hope to shed light on how they decide between – and prioritise – several different courses of action. The results could inform work on human conditions such as Obsessive Compulsive Disorder (OCD).

Meanwhile, the Group’s work with robots is providing useful insights for applications such as computer games and the development of intelligent agents (autonomous computer processes to help people find the information that they want, when they want it). It may also help to find ways of avoiding, or at least coping with, a common problem that affects many people in an increasingly busy world: Dithering, or oscillating between several tasks because we are unable to decide which of them is the most important.

While all these potential spin-offs are valuable, the group’s ultimate goal is even more profound.

‘By combining insights from neuroscience, neural network research and robot modelling we are hoping to understand more about how our brain works,’ says Dr. Prescott. ‘This is the real driving force behind most of our work. Ultimately it’s about understanding ourselves.

Contacts:

For more information e-mail Dr Tony Prescott at t.j.prescott@sheffield.ac.uk

This article was reproduced from EPSRC’s Newsline magazine.