A new species of hardware: Ordinary hardware does the same old job until it
wears out, whereas evolvable hardware adapts itself to a changing task 

By: Moshe Sipper, Swiss Federal Institute of Technology and
Edmund M.A. Ronald, Ecole Polytechnique, Center for Applied Mathematics 

From: IEEE Spectrum - March 2000 - page 62
http://www.spectrum.ieee.org/spectrum/mar00/features/evol.html

Tetsuya Higuchi and his colleagues at the Evolvable Systems Laboratory of the
Electrotechnical Laboratory in Tsukuba, Japan have built a general-purpose
chip for on-line evolvable hardware and used it to implement a controller for
an artificial hand controlled by electric pulses from the nerves in the arm
muscles. In normal conditions, a disabled person trains for over a month
before being able to manipulate a prosthetic hand with ease. Reversing this
scenario, Higuchi and his colleagues had the artificial hand adapt itself to
the disabled person, instead of having the person adapt to the hand. The idea
was that the on-line controller should accept signals from the nerves in the
arm and map them to desired hand actions. Because those signals vary greatly
between individuals, it is impossible to design such a circuit in advance.
But with the evolvable-hardware controller, the hand usually requires less
than 10 minutes to adapt to its owner through on-line training, in which the
person repeats various hand movements - a notable improvement over the one
month required when the owner is the one doing the adapting. 

Picture caption: Researchers at the Electrotechnical Laboratory in Tsukuba,
Japan used evolvable hardware to develop a prototype artificial hand that
adapts to a patient's unique arm-muscle activity. Typically, the adaptation
is the other way arround: people control the hand by learning to modify their
muscle activity. 

http://www.etl.go.jp/etl/divisions/~ehw/eng/noframe/main.html
http://www.etl.go.jp/etl/divisions/~ehw/common/gishu/gishu.html
