Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!newsgate.duke.edu!zombie.ncsc.mil!news.mathworks.com!newsfeed.internetmci.com!in2.uu.net!bcstec!nntp
From: Mike Fouche <mikef@hsvaic.hv.boeing.com>
Subject: Re: Controlling an Inverted Pendulum ?
X-Nntp-Posting-Host: e677744.hv.boeing.com
Content-Type: text/plain; charset=us-ascii
Message-ID: <Ds6pA8.Byy@bcstec.ca.boeing.com>
Sender: nntp@bcstec.ca.boeing.com (NNTP News Access)
Content-Transfer-Encoding: 7bit
Organization: Missiles & Space Division
References: <4o4ld6$hvu@nx2.HRZ.Uni-Dortmund.DE>
Mime-Version: 1.0
Date: Wed, 29 May 1996 20:37:19 GMT
X-Mailer: Mozilla 1.2 (Windows; U; 16bit)
Lines: 21

Actually yes.  But our experience has been that one layer suffices.  A 
typical neural controller that we use will have 4 inputs (theta, 
thetadot, x, xdot), one hidden layer with 4 neurons, and one output 
neuron which generates the force.  These neural controllers were 
generated by training them on the transient response of a closed-loop 
classical controller.  

If you want to shape the response of the system via the neural controller
then 6 inputs are required but you still only need one hidden layer.

I've also designed neural controllers from just the open-loop
response of the system - again they only contain one hidden layer.

I'd suggest you write your own simulation, train your own nets, and see
what happens.  I'll look through my stash to see what useful *detailed*
information there is.
  

Mike Fouche
Boeing Missiles & Space Division

