🚀 go-pugleaf

In article <C0ICnz.o6@world.std.com> lains@world.std.com (Layne L Ainsworth) writes:

   Has anyone tried to do this: 

   "Herds" of cheap (really cheap), reprogrammable robots, each loaded
   with one set of the test "chromosome" population of a Genetic
   Algorithm system. They could be evaluated to optimize indivdual
   performance, or "hive" well-being.  Each "generation" would be the
   same robots, with a new generation of "chromosomes" on board.

   Pointers to papers or discussion of the feasibility of such a project
   would both be welcome. I would think this could be a job for the
   much-heralded and -scorned MiniBoard. Wire up all the connections to
   various sensors/actuators and let the herd discover uses for them.

   My dream, of course, is a bunch of robots that (who?) could
   continuously clean my house. :-)

   Thanks.

   -Layne

I don't know of anyone who has done anything like this with real
robots, but I do know of instances of using genetic models to
program robotic behaviour under simulation.

John Koza's "Genetic Programming: On the programming of computers by
means of natural selection" from MIT press (1992) has examples of this
in simulation.

"Genetic Programming - the Movie" Koza & Rice MIT Press 1992 has
visualisations of the learned robotic behaviour.

(Am I beginning to sound like a broken record?)

As well as these I know that Craig Reynolds <cwr@white.sww.symbolics.com>
has been looking at the genetic evolution of flocking/swarm
behaviour.  Koza's book also covers emergent behaviour of ants
(actually it covers 81 different problem domains, so the GP
question is likely to be a reasonable answer to a wide set
of queries similar to your's.  The movie covers >20 diffeent
problem domains).

There's a student at Stanford, whose name escapes me who's
used GP to program Beer's six legged robot.  If you want
I can doubtless dig up a pointer.

Craig Reynolds had a paper at the Simulation of Adaptive Behaviour
conference in Hawaii last month, but I haven't seen it yet.


As of a couple of days ago, there is now a GP mailing list.
Send mail to Genetic-Programming-Request@CS.Stanford.Edu to
subscribe.



Rice.

This is only sort-of related to GAs for robot behavior but I did
a masters' thesis on using GAs in a prolog system that attempted
to emulate the language learning abilities of bottlenosed dolphins.
We trained the simulation in the morning and trained the real dolphinsj
in the afternoon.  We tried to get the simulation to predict what sort
of things the real dolphins might do.

Brian Tarbox
Been There - Done That Software

In article <C0ICnz.o6@world.std.com> lains@world.std.com (Layne L Ainsworth) writes:
>Has anyone tried to do this: 
>
>"Herds" of cheap (really cheap), reprogrammable robots, each loaded
>with one set of the test "chromosome" population of a Genetic
>Algorithm system. They could be evaluated to optimize indivdual
>performance, or "hive" well-being.  Each "generation" would be the
>same robots, with a new generation of "chromosomes" on board.

Larry Yaegar (sorry if I butchered the last name!) at apple (larryy@apple.com
is his address if memory serves) has a wonderful simulation that runs on
SGI machines.  It falls most squarely under the category of Artificial Life,
since the evolving creatures are implicitly rewarded for survival and 
reproduction (i.e., they must feed, defend themselves, and mate while others
around them do the same) rather than being rewarded for an explicit externally
imposed fitness function.  Different genotypes evolve, each with their own
niche.  Most go extinct fairly rapidly, some flourish.  Very visual, this work
will probably be published in the forthcoming Artificial Life III procedings
and the software can be obtained from Larry Yaegar (I think).

On a less visual (but more supervised) note John Koza and James Rice have
done work with emergent behavior of ants (see ALife-2 proceedings & video, and
 JK's new book "Genetic Programming," and the accompanying video by Koza & 
Rice.).  Also see work by Collins & Jefferson (also in ALife3) and Rob Collins
dissertation, available via ftp from UCLA/cognet.

-walter alden tackett

Disclaimer: "My employers are responsible for the contents of this letter 
             because I secretly control their thoughts."

My Co., Global Associates, is doing work in this area for DARPA.  We are
using collective behavior in robots to survey an area.  The Behavior isn't
necessarily evolved with a GA but the ant analogy is very close.  

-Richard Freytag

In article <C0IK85.C51@world.std.com> btarbox@world.std.com (Brian J Tarbox) writes:
>We trained the simulation in the morning and trained the real dolphinsj
>in the afternoon.  We tried to get the simulation to predict what sort
>of things the real dolphins might do.

So don't leave us hanging -- what were the results?
-- 
Tom Atwater

In article <C0ICnz.o6@world.std.com> lains@world.std.com (Layne L Ainsworth) writes:
>
>
>Has anyone tried to do this: 
>
>"Herds" of cheap (really cheap), reprogrammable robots, each loaded
>with one set of the test "chromosome" population of a Genetic
>Algorithm system. They could be evaluated to optimize indivdual
>performance, or "hive" well-being.  Each "generation" would be the
>same robots, with a new generation of "chromosomes" on board.
>
>Pointers to papers or discussion of the feasibility of such a project
>would both be welcome. I would think this could be a job for the
>much-heralded and -scorned MiniBoard. Wire up all the connections to
>various sensors/actuators and let the herd discover uses for them.
>
>My dream, of course, is a bunch of robots that (who?) could
>continuously clean my house. :-)
>
>Thanks.
>
>-Layne
>
>
>
>-- 
>  Layne Ainsworth | lains@world.std.com <or> ...!{xylogics,uunet}!world!lains
>	  The truth can't hurt you; it's just like the dark.
>



Check out

        "The Learning of Reactive Control Parameters through Genetic
         Algortihms", M. Pearce, R. Arkin, and A. Ram, Proc. 1992
         IEEE/RSJ International Conference on Intelligent Robots and
         Systems, (IROS 92), Raleigh, NC July 1992, pp. 130-137.

These are the simulation results, we hope to port these to our Denning
Robots shortly.

- Ron
   
Ronald C. Arkin
College of Computing, Georgia Institute of Technology
Atlanta, Georgia 30332-0280
Phone: (404)-894-8209            Fax: (404)-853-9378

In <24659@hacgate.SCG.HAC.COM> tackett@ipld01.hac.com (Walter Alden Tackett) writes:

>Larry Yaegar (sorry if I butchered the last name!) at apple (larryy@apple.com
>is his address if memory serves) has a wonderful simulation that runs on
>SGI machines.  It falls most squarely under the category of Artificial Life,
>since the evolving creatures are implicitly rewarded for survival and 
>reproduction (i.e., they must feed, defend themselves, and mate while others
>around them do the same) rather than being rewarded for an explicit externally
>imposed fitness function.  Different genotypes evolve, each with their own
>niche.  Most go extinct fairly rapidly, some flourish.  Very visual, this work
>will probably be published in the forthcoming Artificial Life III procedings
>and the software can be obtained from Larry Yaegar (I think).

I think it's also available via anonymous ftp at "ftp.apple.com" and maybe
at "ftp.cognet.ucla.edu" as well.

Cheers,
Rob




-- 
[--------------------------------------------------]
[  Robert S. Mah   |  "Every day an adventure,     ]
[  rmah@panix.com  |   every moment a challenge."  ]
[--------------------------------------------------]

One more addendum to this topic: if you are interested in evolving the behavior
of single robots, I received a couple of papers from the U. of Sussex at
Brighton related to this topic.  As far as I know they are internal tech
reports:

``Issues in Evolutionary Robotics,'' I. Harvey, P. Husbands, and D. Cliff
CSRP #219, July 1992

``Evolving Visually Guided Robots,'' D. Cliff, P. Husbands, and I. Harvey
CSRP #220, July 1992

Authors may be contacted at: davec philh and inmanh @cogs.susx.ac.uk

-walter alden tackett

`` If we do not succeed, then we run the risk of failure. ''
			- (almost former) U.S.V.P. Dan Quayle

In article <24659@hacgate.SCG.HAC.COM> tackett@ipld01.hac.com (Walter Alden Tackett) writes:
} In article <C0ICnz.o6@world.std.com> lains@world.std.com (Layne L Ainsworth) writes:
} 
} Larry Yaegar (sorry if I butchered the last name!) at apple (larryy@apple.com
} is his address if memory serves) has a wonderful simulation that runs on
} SGI machines.  It falls most squarely under the category of Artificial Life,
} since the evolving creatures are implicitly rewarded for survival and 
} reproduction (i.e., they must feed, defend themselves, and mate while others
} around them do the same) rather than being rewarded for an explicit externally
} imposed fitness function.  Different genotypes evolve, each with their own
} niche.  Most go extinct fairly rapidly, some flourish.  Very visual, this work
} will probably be published in the forthcoming Artificial Life III procedings
} and the software can be obtained from Larry Yaegar (I think).

I was part of a group that did a project very much like at GWU a while back,
though ours was quite non-visual.

"Survival of the Fittest".  G. Bailey, R. Demeo, H. Holz, and J. Revell.
Proceedings of the Workshop on Evolution and Chaos in Cognitive Processing
(IJCAI-91), August '91.

There is additional work, but it's not particularly related to the original
request.
-- 
James Revell	Network Services Mgr   <revell@uunet.uu.net>   /8^{~

In article <C0ICnz.o6@world.std.com> lains@world.std.com (Layne L Ainsworth) writes:
>
>My dream, of course, is a bunch of robots that (who?) could
>continuously clean my house. :-)
>

Of humans?

Steve  :)