Newsgroups: comp.ai.fuzzy,comp.soft-sys.matlab
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!uhog.mit.edu!news.mtholyoke.edu!world!borg
From: borg@world.std.com (Craig W Borghesani)
Subject: Re: PWS_Matlab_Fuzzy_Toolbox
Message-ID: <CwqpvD.3Dn@world.std.com>
Followup-To: comp.ai.fuzzy,comp.soft-sys.matlab
Organization: The World Public Access UNIX, Brookline, MA
X-Newsreader: TIN [version 1.2 PL2]
References: <Cw83nI.4nC@cee.hw.ac.uk>
Date: Mon, 26 Sep 1994 14:06:48 GMT
Lines: 109
Xref: glinda.oz.cs.cmu.edu comp.ai.fuzzy:2983 comp.soft-sys.matlab:8524

Donald Reay (dsr@cee.hw.ac.uk) wrote:

: I've just downloaded the demonstrations of the PWS FUZZY SYSTEMS TOOLBOX and
: have come up with a suggestion for a modification to the truck backer demo.

: It seems a bit uninteresting to compare the fuzzy controlled truck with an
: UNCONTROLLED truck. Who expects an uncontrolled truck to back up?

: The problem tackled seems to be to drive the truck angle to pi radians and
: the horizontal component of the truck position to 50 metres. In order to
: achieve the former, steer in the sense (i.e. +ve or -ve) of the truck angle
: error. In order to achieve the latter, steer on the opposite sense to the
: truck horizontal position error. (This from a common sense understanding of
: a simple truck, which seems to be what has been simulated.)

: Therefore, why not make the steering angle a linear combination of angle
: and horizontal position errors? Better limit the magnitude of the applied
: steering angle to simulate full lock.

: To see what I mean, replace the line
:  
:   u = frule(A,C,Z,[s(1); s(3)]);

: in the file trkdemoe.m (this is the PC version) with
: ---------------------------------------------------------- 
:   % replace fuzzy controller with linear controller    
:   %
:   % hard limit applied steering angle to this magnitude
:   % even if linear controller demands more
:   maxsteerangle = 0.35;  
:   % controller gains to act on position error (m)
:   % and truck angle error (radians) respectively
:   positiongain = 0.03;
:   anglegain = 0.3;
:   % linear controller  
:   u = (positiongain*(s(1)-50) - anglegain*(s(3)-pi));  
:   % limit controller output  
:   if u > maxsteerangle
:     u = maxsteerangle;
:   end
:   if u < -maxsteerangle
:     u = -maxsteerangle;
:   end
: ----------------------------------------------------------

: and run fuzzdemo, choosing controller 1 for the truck.

: I've tried 0.01 < positiongain < 0.04, 0.1 < anglegain < 0.4,
: and maxsteerangle = 0.35

: I deduced the gains and full lock from the response plots for fuzzy
: controller 1, i.e. by observing one successful back up. (Bear in mind my  
: earlier characterisation of the problem - I'm not expecting extremely
: non-linear behaviour from the truck.)

: OK, my control design method may not bear close scrutiny regarding stability
: or robustness but it gives something else to compare the performance of the
: fuzzy controller with.

: Finally, isn't the truck simulated here rather simpler than the one used
: by Nguyen and Widrow? I realise that the FUZZY SYSTEM TOOLBOX authors do not
: claim that their truck is the same - I'm just asking.

: Any comments?

: Anyone had any experiences with the FUZZY TOOLBOX itself?

: Donald

a friend of mine is unable to post to this newsgroup from where he is, so 
he asked me to do it.

-----------------------

This is in response to an earlier posting (by Donald Reay )
regarding the truck backer-upper demo in the Fuzzy Systems Toolbox.

>It seems a bit uninteresting to compare the fuzzy controlled truck with an
>UNCONTROLLED truck. Who expects an uncontrolled truck to back up?

The simulation of the uncontrolled truck is there simply to
show the system that is to be controlled.  I agree its not
useful as a benchmark for controller performance. Although, I
have seen instances of human truck backing which might suffer
from the comparison. :)

>Therefore, why not make the steering angle a linear combination of angle
>and horizontal position errors? Better limit the magnitude of the applied
>steering angle to simulate full lock.

>Finally, isn't the truck simulated here rather simpler than the one used
>by Nguyen and Widrow? I realise that the FUZZY SYSTEM TOOLBOX authors do not
>claim that their truck is the same - I'm just asking.

There are many simple solutions to both the FST problem and the
Nguyen-Widrow problem.  A solution similar to the one you suggest
has been published by an Austrailian (I don't have the reference
with me) which solves the problem for trucks with up to two trailers.

On the other hand, the fuzzy system solution does better than
saturating linear solutions when the truck is started close to
the bay, and is more suitable for extending to a more complex system.

Mark Beale





