M.Yamakita

1
Introduction To Intelligent Control

• M.Yamakita
• Dept. of Mechanical and Control Systems Eng.
• Tokyo Inst. Of Tech.

2

• Controlled system becomes more and more complex.

• It is almost impossible to represent mathematical
  differential and difference equation
  representation of the systems.

Emergent technology is needed !
Intelligent Control
3
Trends from 60
Artificial Intelligence (AI)
Crisp Logic
Fuzzy Logic
Symbolic Representation
Non-Symbolic Representation (ANN)
Control Theory
Classical Control Theory (PID)
Modern Control Theory Robust Control
Theory Adaptive Control Theory Hybrid System
Control

4
Whats Intelligent Control ?
Intelligent Control
(
Intelligent Control ? Fuzzy Control
5
Structure Of Intelligent Control
1. Hierarchical Intelligent Control (Albus,
Saridis)
2. Reactive Intelligent Control (Brooks)
(Subsumption Architecture)
6
Hierarchical Intelligent Control (Saridis)
PRECISION
INTELLIGENCE
7
Reactive Intelligent Control (Brooks)
Modify the World
Create Maps
Discover New Area
Avoid Collisions
Move Around
8
Supporting Technologies
1. Extensions of conventional control technologies
Robust optimal control
Adaptive control
Learning control
2. New technologies
FAN(Fuzzy, AI, and Neural network) technology
(Fukuda)
Soft computing (Zadeh)
9
Dynamical System Representation(State Space
Representation)
10
Robust Optimal Control
Set of uncertain systems
A nominal system
Model set of uncertain systems
11
Adaptive Control
12
Symbolic System Representation(Rule Based
Representation)
Area3
Area2
?
Area1
Classical AI, Automaton etc.
13
Crisp Logic vs. Fuzzy Logic
Tall
Mr.A
180cm
170cm
?
Mrs.B
(
170cm
160cm
Mr.C
Short
14
When we describe real world symbolically, there
always exist gray zone state. It is very
difficult to describe the gray zone property by
conventional crisp logic. Or, we must define
undesirably many categories.
Fuzzy Logic
15
Introduction of membership functions
Degree of property
100
50
)
x
160
180
170
Height
Short
Tall
16
Perceptron
O1
O2
B
B
A
C
A
C
D
A--B B--C C--D
A--B (A is connected to B) B--C C--A
Triangle
NOT Triangle
Human easily recognize O2 as triangle !
17
Mimic the brain function !
18
No Hidden Layer
(Rosenbratto Type Perceptron)
19
Multi Layered Neural Network
Adjustable Weights
Activation Function
. .
Generalized delta rule, Back-propagation algorithm
(Amari, Rumelhalt)
20
References

• 1.M.M.Gupta,N.k.SinhaIntelligent Control
  Systems,
• IEEE Press. (1996)
• 2. K.Furuta et.Intelligent Control, Corona Pub.
  (1988)
• (in Japanese)
• 3.B.Widrow, E.Walach Adaptive Inverse
  Control,Prentice Hall (1996)