Newsgroups: comp.robotics
Path: utzoo!utgpu!watserv1!watmath!mwtilden
From: mwtilden@watmath.waterloo.edu (M.W.Tilden, Hardware)
Subject: Re: CM Ambler Rover
Message-ID: <1990Jun4.160147.24319@watmath.waterloo.edu>
Organization: University of Waterloo
References: <3708@jato.Jpl.Nasa.Gov> <95816@philabs.Philips.Com> <BBC.90May19164725@sicilia.rice.edu> <1990May30.182249.22352@watmath.waterloo.edu> <18280@well.sf.ca.us>
Distribution: na
Date: Mon, 4 Jun 90 16:01:47 GMT
Lines: 61

In article <18280@well.sf.ca.us> nagle@well.sf.ca.us (John Nagle) writes:
>In article <1990May30.182249.22352@watmath.waterloo.edu> mwtilden@watmath.waterloo.edu (M.W.Tilden, Hardware) writes:
>>In article <BBC.90May19164725@sicilia.rice.edu> Benjamin Chase <bbc@rice.edu> writes:
>>Brooks at MIT seems to have a reasonable solution.  They've built a 
>>6 legged walker which uses a simple Algorithmic State Machine network
>>as the control (simulated in a single 8bit 68HC11 processor).  The 
>>thing they built is the size of a cat but walks and thinks very
>>much the way any 6 legged bio-critter does.  
>
>      The last above is an assumption probably not justified by the
>facts.  But see "A Biological Perspective on Autonomous Agent Design",
>by Beer, Cheil, and Sterling, at CWRU, Cleveland, OH. (which may have 
>appeared in print by now; I have a copy I got at MIT.)  These people
>have a six-legged walking simulator which is based on an explicitly
>biological neural model.  The model Brooks uses isn't particularly
>representative of any specific biological model.  Patty Maes has a
>technique by which Brooks' machine is made to learn to walk, but her
>approach isn't something one would expect from a biological system
>either.  (It may be better; it learns to walk in a minute or so.)

Sorry if I brought that idea across, but if you saw the evolutional 
stages of this device dragging it's bum in an attempt to climb over 
a telephone book, then you would be immediately reminded of a kitten 
trying to climb stairs.  The similarity was so strong I burst out 
laughing during the one lecture I attended.  I was not alone.

The flyer I have from that speech given by Brooks sez "rather, like children
or dogs, they do what is in their nature (...determined by [LEPROM 
programming] ...residing inboard...)"  Brooks made no further biological
references.  My talks to Biochemists and Biologists however seemed to give
me the idea that "yeah, that's pretty close to the mark.  Needs heuristics
however."

It makes me wonder though if, as fractal mountians resemble real ones
even though formed by completely different processes (recursion/
erosion), can complex biological behaviours be successfully mimicked by 
very simple FSM (Finite State Machine) interconnections?  Brooks
technique is to bundle his beasts with sensors and have all of them
contribute to the creatures reactions.  The creature has no memory of what
it has done, only reactions to what is happening now. However, his 
creatures do not learn from their mistakes, their programmers do.  
I asked him why he had not thought of building a learning ability 
into his critters and he said "Not necessary.  The turn around time 
for programming [the walker] is far faster than any I've seen for 
similar devices.  More effective too."

Good enough for me.  I've already got a bitty-critter made which I'm
going to attempt operation using the same technique.  I'll post results.

For those who may be looking for references, the mans name is 
Rodney A. Brooks of the MIT Artificial Intelligence Lab.  If anybody
has exact references to papers on this topic, a post would be appreciated.

Is all.


-- 
Mark Tilden: _-_-_-__--__--_      /(glitch!)  M.F.C.F Hardware Design Lab.
-_-___       |              \  /\/            U of Waterloo. Ont. Can, N2L-3G1
     |__-_-_-|               \/               (519) - 885 - 1211 ext.2454,
"MY OPINIONS, YOU HEAR!? MINE! MINE! MINE! MINE! MINE! AH HAHAHAHAHAHAHAHAHA!!"
