In recent years, the application of modern control theory is rapidly expanded to the area of the adaptation and learning problems, with the developement of computer techniques.
Although theoretical discussions for these problems are very active, it seems that there are not enough considerations and discussions about the technical realizations or practical applications of those problems. Especially, as data processing devices, it is not absolutely necessary to use the expensive universal computers only, and it is expected that some cheaper simple devices are available in some parts of control subjects. It is necessary to attack such investigations which have been little interested, and it is disired to make the possibility of the practical realizations of the theory apparent in some degrees.
Consequently, this paper is intended to show the possibility of the realization of Kalman filtering by constructing a rather simple data processing device.
This is for a linear learning control system corrupted by additive gaussian noises with unknown means, for which the authors have synthesized the optimal control policies.
This device consists of simple conbinations of elements, ie relays, analogue-computers, digitalvoltmeters, potentiometers, punch-tapes, tape-readers and so on.
The experimental results on this device are discussed in the case of an one-dimensional control system.
Some discussions for the improvements of this device (multidimensionalization, speed-up of operations, an improvement of calculation accuracy and so on) are presented.
View full abstract