Neuroscientists at Sussex University have made a world-first discovery which could revolutionise the science of robotics. In the hope of creating robots which are completely autonomous, Dr Phil Husbands and Professor Michael O'Shea have transformed the workings of a robot's brain - by unleashing a toxic gas.

The gas - nitric oxide - is actually integral to the neurology of the human brain. Our bodies use nitric oxide to regulate blood flow, but it is also used in the brain to allow neurons to communicate with each other. The neurons produce the gas, and as the gas diffuses it enables these neurons to send messages to other neurons with which they are not physically or electrically connected. The discovery of nitric oxide as a message-carrying gas was, according to Professor O'Shea, "a completely new way of thinking about how the brain works." Adapting it for use in a robot's brain is even more radical.

Until now, attempts to replicate the human brain in robots have been hampered by the impossibility of creating something which, like our brains, has 100 billion neurons which connect with each other in a huge electronic mesh. Attempts to improve robot's brains have so far concentrated on trying to make as many 'neurons' as possible, which are as connected to each other as possible. The Sussex team - Professor O'Shea, Dr Husbands and their students Tom Smith and Andy Philippides - have developed a brain which by-passes this elaborate process almost entirely, by simulating the action of the nitric oxide in our brains. As a result, Michael O'Shea claims, "it transforms robots' brains from being pretty slow and stupid to being faster, more adaptable and clever."

"The gas allows messenger molecules to function over a much broader spectrum in the human brain," says Professor O'Shea, "and, using a virtual gas, we can replicate this process in robots' brains." Using a mathematical construct, the existence of the gas is plotted into the computer which develops the robot. The robot is then allowed to 'evolve', with the virtual gas a central part of this 'evolutionary' process.

The Centre for Computational Neuroscience and Robotics, co-directed by Michael O'Shea and Phil Husbands, was a pioneer in the concept of evolution as a way of developing robotic technology. As Professor O'Shea points out, the process derives its inspiration from natural biology - "we incorporate biological reality into artificial systems and computers."

Robots which have 'evolved' haven't actually mated with other robots - a task has been set, and a computer programme run which takes the representative of each generation which seems most capable of the task and puts them forward into the next generation, where the most capable individual will again be put forward into the next generation and so on. At the moment, most tasks require an evolutionary cycle of around 6,000 generations before a robot is capable of the set task. "By incorporating gases into the programme, you can solve an equivalent task in ten times fewer, and sometimes 100 times fewer generations," Michael O'Shea says.

The brains of the new robots developed at the CCNR also have a much simpler internal system, because the gas cuts down the need for simulated neurons and the complex connections between them. This means that the process of their evolution, and indeed everything that determines how and why they work, becomes much more open to analysis. According to Professor O'Shea, "The exciting opportunity is that we could then solve task using gasnets that would be impossible with conventional neural networks."

At the moment, robots are only capable of relatively simple tasks. A robot which can construct a car on a production line would not be able to vacuum a child's bedroom, because such a task requires a level of intelligence neuroscientists have come nowhere near to developing. As Michael O'Shea points out, "To clean a child's bedroom, a robot would have to be so intelligent, it would virtually be as intelligent as a human being, and then it would probably tell you to do it!" However, with the introduction of gasnets, "we're actually producing a robot which demonstrates quite sophisticated behaviour, as well as being almost autonomous."

The CCNR robot may not be able to vacuum a child's bedroom - yet. But, as Michael O' Shea points out, "Intelligence really springs from interaction with an unpredictable and unforgiving world," and robots using gasnets may be able to perform tasks in the near future which do require a measure of this intelligence. Tasks like clearing minefields, or helping with space missions, all cause problems for humans which would not affect robots in the same way. C3PO may become the new symbol of humanitarian aid.

For further information please contact Sally Hall, Information Office, University of Sussex, Tel. 01273 678888, Fax 01273 678335, email s.l.hall@sussex.ac.uk, Professor Michael O'Shea, Centre for Computational Neuroscience and Robotics, University of Sussex, Tel. 01273 678055, email m.o-shea@sussex.ac.uk, or Dr Phil Husbands, Centre for Computational Neuroscience and Robotics, University of Sussex, Tel. 01273 678556, email p.husbands@sussex.ac.uk.