No flies on Sussex robotic scientists

The normal response to the buzz of two flies forever chasing each other around a lamp is to swat them. But for researchers at Sussex University the flies' behaviour is a source of amazement. It indicates just how complex and ingenious the brains of even the most simple living systems can be.

Michael O'Shea and his colleagues at the university's interdisciplinary research centre for neuroscience believe that by understanding the behaviour and neurology of simple animals such as houseflies, snails and locusts, big advances may be possible in the design of novel computer systems and autonomous robots.

Their work recently received a big boost with a Pounds 300,000 award by the university to the IRC to help it set up a centre for computational neuroscience and robotics. The initiative will link biologists at the IRC, which is allied to the Biotechnology and Biological Sciences Research Council, with computing experts at Sussex's school of cognitive and computing sciences.

Dr O'Shea, director of the IRC, said: "We want to derive inspiration from biological principles to help us design software and hardware solutions that so far have proved difficult to come by using conventional computational methods."

He says that the initiative has already attracted the attention of firms. British Telecom is interested in the application of biological principles to the design of information processing systems.

Biologists hope to learn a great deal about how nervous systems work by computer modelling those already well researched. This will allow them to do "virtual" experiments on well defined neural networks - experiments that are often impossible to do directly on the animal.

The IRC's deputy director Paul Benjamin and PhD student Volko Straub are researching the brain cells responsible for the rhythmic feeding behaviour of snails. A biological understanding of the neuronal networks involved when a snail eats is helping to create computer models that mimic the networks. Professor Benjamin hopes that the work will provide powerful insights into how the snail brain generates simple patterns of behaviour and how information is stored in their neural networks.

Meanwhile the team of computer scientists hopes that the collaboration with biologists will help them understand how evolution has solved most of the hardware and software problems that perplex them. They plan to apply solutions developed by simple animals to machine intelligence and autonomous robots. One day such robots may approach, perhaps even match, the ability of animals and humans to memorise, learn, navigate, move and respond to the aural and visual world.

The computing team, led by Phil Husbands and David Cliff, is already developing genetic algorithms to generate computer programmes for robots. The technique borrows from the theory of evolution by natural selection, which argues that populations contain natural variations due to the occurrence of mutations. An iterative selection process operating on that variation results in new generations possessing new, and often improved characteristics.

Applying this process of Darwinian natural selection to the design of computer programmes offers the possibility of creating effective artificial robot systems that "breed" over thousands of generations to perform certain tasks.

Simple animals manage to perform extraordinary feats, such as seeing and navigating, using very few neurons. Houseflies have brains that are no bigger than a pinhead and yet they are capable of extraordinarily rapid responses. Michael Land, an expert on animal and human vision based at the IRC, explains that two flies chasing each other typically travel at two to three metres per second. Their reaction time, for example the time taken by a fly to change direction in response to another fly's move, is about 15 milliseconds.

Detailed analysis of this behaviour by Professor Land led him to propose a description of the neural architecture needed by a fly in order for it to respond in this manner. The prediction was subsequently confirmed as being neurologically correct.

Professor Land recently received a Pounds 75,000 prize from the United States pharmaceutical giant Alcon Laboratories for outstanding work in ophthalmology. Projects under way in his laboratory include analysing human eye movements during driving, playing the piano and playing table tennis.

He has devised a unique camera headset that monitors a person's eye movements while simultaneously recording their view and direction of gaze. By placing the headset on drivers, he has discovered that all drivers, regardless of speed of travel, fix their gaze at a particular point on bends as they drive around them.

Such work is helping him to develop his ideas about the kind of neural architecture and information processing strategies the human brain might be using under such circumstances - ideas that can be fed into the work of the new research centre.

But it also has more immediate value: "Identifying the visual processes involved in acquiring these skills should suggest strategies for novices to learn to drive more effectively as well as indicating where road signs should be located for maximum safety."