SPACE explorers of the future won't be brave heroes like Yuri Gagarin and Neil Armstrong-they are more likely to be autonomous robots capable of thinking, learning and seeking out new environments. So say researchers at electronics company Raytheon Systems and the University of Texas at Dallas.
The researchers have developed an electronic neural switch for an artificial nervous system (ANS) they are planning. In about a decade they hope the electronic system will mimic the actions of the human brain and its extended communications network, and so make possible an autonomous robot that can take in information from many sensors and make its own decisions.
"What we've done is take a biological approach to building such an autonomous machine," says principal researcher Larry Cauller, a neuroscientist at the University of Texas. In a biological nervous system, each neuron interacts with thousands of others. But today's computer processors can only be linked up in small groups before control software becomes unwieldy, explains his colleague Andrew Penz of Raytheon.
To create complex brain functions, the team needs to mimic actual neural activity better. So Penz and his colleagues have designed nanoelectronic processors which they say can act like neurons. These gallium arsenide processors have layers thin enough to allow electrons to fan out or "tunnel" in many directions, making each electron stream behave as if it were inside a one-way device such as a diode.
Increasing numbers of electrons pass through the same channel until their motion becomes inhibited by their own electric fields. Electron numbers then drop until the field dissipates; then the process starts up all over again. This tunnelling pattern forms an N-shaped curve, which the team says is remarkably similar to that displayed by the sodium and potassium ion channels in nerve cells.
Cauller and Penz's computer simulations have shown that devices like this could transmit signals similar to a pulse from a nerve cell. By placing thousands of these nano-sized processors, called resonant tunnelling diodes (RTDs), on one chip, Cauller expects to be able to build an artificial brain. The researchers have written a simulation of the behaviour of such a chip which has at its centre a "cortex" containing six layers of virtual interconnecting RTDs. The cortex is linked within the simulation to a surrounding core of devices that control robotic actuators which are in turn linked to environmental sensors. The connections between the cortex and the environment are two-way. This allows the brain to observe the results of its own actions, potentially giving it the power to learn and anticipate. It will be much faster than today's software-based neural networks because it is based on "hardware neurons". "We hope that it will start learning from environmental stimuli early next spring," Penz says.
The ANS research was initially part of a military-sponsored project, but the Defense Advanced Research Projects Agency has lost interest in the robot as a military application and has cut its funding. Says Cauller: "I can't imagine a brain or automaton that searches and learns from its environment becoming a soldier. I prefer to see it as an explorer. You could even hook one of these brains up to the Internet and let it explore a virtual medium."
PLEASE MENTION NEW SCIENTIST IF YOU USE THIS ARTICLE - THANK YOU
UK CONTACT - Claire Bowles, New Scientist Press Office, London:
Tel: 44-171-331-2751 or email claire.bowles@rbi.co.uk
US CONTACT - Barbara Thurlow, New Scientist Washington office:
Tel: 202-452-1178 or email newscidc@idt.net
Journal
The New Scientist