Sammendrag
This thesis is about the use of Evolutionary
Algorithms to design a better prosthetic hand controller. One of the
goals is to use methods that are easy to implement in a small, low-power
and low-cost system. The data set used is typical of the data that
would be available in a real-world prosthesis. It was collected by
Kajitani at the National Institute of Advanced Industrial Science
and Technology (AIST) from a person who had lost a hand, and no advanced
preprocessing of the signal was done. Evolutionary Algorithms are
used to evolve a digital circuit which can predict the intended hand
motion from the data presented to it. The data set is then analyzed
to determine the factors that limit the successful classification
of signals. The maximum classification rate attainable is determined,
and the expected maximum real-word performance is also evaluated.
Finally, a method is found that improves the average classification
rate at the cost of increased response time. Compared to another work
using the same data set, the average classification rate for the testing
data rose from 55.1\% to 71.2\%, for the training data it rose from
73.1\% to 92.3\%.