1990 Fiscal Year Final Research Report Summary
NeuroーController with the Learning Function
| Project/Area Number |
01550311
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
電子機器工学
|
|---|
| Research Institution | Osaka University |
|---|
Principal Investigator |
TAGUCHI Hideo Osaka Univ., Fac. of Eng., Lecturer, 工学部, 講師 (40029278)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
AKAZAWA Kenzo Kobe Univ., Fac. of Eng., Professor, 工学部, 教授 (30029277)
AKAZAWA Kenzo Kobe Univ., Fac. of Eng,. Professor (30029277)
AKAZAWA Kenzo Kobe univ., Fac. of Eng., Professor (30029277)
AKAZAWA Kenzo Kobe Univ,. Fac. of Eng., Professor (30029277)
AKAZAWA Kenzo Kobe Univ,. Fac. of Eng., Professor (30029277)
|
|---|
| Project Period (FY) |
1989 – 1990
|
| Keywords | Neurocomputing / Artificial Neuron / Pulse Operating / Neurofilter / Neurocontroller / Neural Servo / Learning Operation / Analog-MOS Circuit |
|---|
| Research Abstract |
In this research, we advanced the development of a neurocontroller by using the nerocomputing technique for the controller design. And, the goalーoriented architecture of the enrocontroller was discussed on both sides of the software and the hardware. After all, the results of this research are summarized as follows :
1. A neuronーlike learning element was formulated in the light of excellent signal processing characteristics of a single bioneuron ; i, e., the V/F conversion, the pulse frequency modulation and the plastic signal transfer function. And also, it was confirmed that the combination of several neuron-like learning elements led to such the successful control operation as lowpass, highpass and bandpass neurofilter.
2. By utilizing variable characteristics of neuron-like learning elements, We built up the neurocontroller that was consisted of neurosensors, PTM (Pulse Time Modulation) type neurofilters and neuroseleclor. Here, neurosensors functioned so as to convert sensory information into pulse frequencies and take in the control system. Each of PTM type nero-filters played the part of any different control characteristics. And, the neuroselector was designed to use properly each of PTM type neurofilters correspondingly to variable control situations. Also, to estimate the responses of neural servo-system with this neurocontroller, a torque control of the single joint manipulator equipped with spring and dash-pot was attempted.
3. To aim at the hardware of the neuro-controller, we designed the mimic electrical circuit to neuron-like learning elements. This mimic circuit was experimentally producted by an analogーMOS device was esperimen tally Produced. And, it was confirmed that the fruitful learning responses to the inputpulse sequences was obtained. Consequently, the possibility to neuroーchip development would be the most necessary for light/miniaturization of the neuro-controller was demonstrated.
|
