Newsgroups: comp.ai.genetic,comp.ai.fuzzy
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!news.duke.edu!news-feed-1.peachnet.edu!gatech!howland.reston.ans.net!EU.net!uknet!comlab.ox.ac.uk!don
From: don@robots.ox.ac.uk (Don Leitch)
Subject: Papers available by ftp
Message-ID: <1994Aug31.150830.16424@hamlet.robots.ox.ac.uk>
Originator: don@hamlet.robots
Organization: Robotics Research Group, Engineering Science Dept, Oxford, UK.
Date: Wed, 31 Aug 1994 15:08:30 GMT
Lines: 91
Xref: glinda.oz.cs.cmu.edu comp.ai.genetic:3783 comp.ai.fuzzy:2784


The following papers are available by anon ftp:

ftp.robots.ox.ac.uk:/pub/ftp/outgoing/don/ 

don_GA94.ps.Z

Context dependent Coding in Genetic Algorithms for the design of Fuzzy
Systems, D. Leitch and P Probert. In: Proc IEEE/Nagoya Univ. WWW on FL and
GA/NN, Nagoya, 1994.

Abstract - 
The use of fuzzy systems for control and robotics is becoming
increasingly widespread, but progress is often hampered by the
difficulty of translating expert knowledge in to fuzzy sets. This
report presents a design approach using 
a genetic algorithm to optimise the rule bases used in two simple
fuzzy
systems. 
We present a novel coding scheme for describing fuzzy systems, in
which the meaning of a section of chromosome is determined by
surrounding genes. This allows any chromosome to be decoded to produce
a fuzzy system, regardless of length or genetic makeup. The genetic
variation of a population is therefore greatly increased, reducing
premature convergence and avoiding the need for complex crossover
procedures. 
The system needs no sample data, and so avoids
dependency on a training data set that can affect fuzzy
clustering algorithms and neural networks. The system learns in an
unsupervised manner, with data being derived from a simulation of the
controlled system. Simulation can cause difficulties, with poor or
ideal simulations of non-ideal situations
leading to solutions that perform poorly in the real world, but
this can be avoided, and in fact the performance of the simulated
system enhanced, by the inclusion of noise in the simulation.
Simulating noise leads to greater exploration of state space by the
genetic algorithm and to improved robustness of the controllers
produced. 


don-eufit94.ps.Z

Genetic Algorithms for the Development of Fuzzy Controllers for
Autonomous Guided Vehicles, D. Leitch and P. Probert. In: Proc. 2nd
European Congress on Intelligent Techniques and Soft Computing (EUFIT
'94), Aachen, 1994.

Abstract - 
This paper proposes an evolutionary technique for generating fuzzy
rule bases for AGV control. The genetic approach to fuzzy rule base
design is outlined, and its applicability to mobile robotics is
discussed. We show that the design problem can be viewed as minimising
a double integral of a complex function of the
vehicle dynamics, sensor configuration, environment and controller.
The suitability of fuzzy rule bases for representing arbitrary control
surfaces is briefly discussed, as is the utility of genetic algorithms
for
minimising the integral by modifying the controller function. We
examine the concept of behaviour in our framework, and how it can be
specified. Finally we present some preliminary results of the
application of the algorithm to some problems in mobile robotics,
namely simple corridor tracking, and performing a three point turn in
a
narrow corridor.


Instructions:

ftp ftp.robots.ox.ac.uk

cd /pub/ftp/outgoing/don

binary

get don_GA94.ps.Z   or   don-eufit94.ps.Z

bye

uncompress don_GA.ps    or   don-eufit94.ps

print using whatever commands you use for printing postscript.


Don.



-- 
Donald Leitch (don@robots.ox.ac.uk)
Research Student, Robotics Research Group, University of Oxford.
19 Parks Rd, Oxford OX1 3PJ, UK. Tel: +44 865 273919 Fax: +44 865 2273908
