Alt-BEAM Archive

Message #03336



To: aubois@trail.com
From: "Jason -" evenflow88@hotmail.com
Date: Tue, 18 May 1999 07:52:52 GMT
Subject: [alt-beam] Re: Zozzles





>From: Jean auBois
>To: "NSX -"
>Subject: Re: Potting (was: Re: Free-forming (was: Contest (mailing robots
>in)))
>Date: Mon, 17 May 1999 13:01:07 -0600
>
>At 04:42 AM 5/1/99 , you wrote:
> >
> >Hi there,
> >can you explain and teach me how I could influence the
> >microcore/bicore with the neural layer/net or some other adaptive sensory
> >inputs?
>
>My first reaction is that I probably don't have much more of a clue than
>you do how to do this. My next reaction (which is probably the most
>useful) is to ask:
>
> (a) What do you want the robot to do / what is the desired behavior?
> (b) Why do you want it to do that?
>
>In other words, you've got some solutions (neural layer/net or adaptive
>something-or-other) and you are looking for a problem to use them on.
>
>Very generally speaking, people tend to use neural nets to learn something
>(the net gets trained) and then, as a result, produce what are supposedly
>correct output given a certain input. These things are not usually built
>from physical representations of neurons -- more often, they are simulated
>in a computer and even then you are talking about needing at least a
>moderate level of skill to get anywhere. If you want to know more about
>doing such a thing, a page on Terry Newton's site:
>
> http://mh105.infi.net/~wtnewton/otherwld/robot4.html
>
>He has done more with putting something "on top of" a 'normal' nervous net
>than anyone else I know.
>
>It is possible, of course, to set up a fixed "neural layer" but this is
>subject to several considerations. The first is: if you know what the
>answer needs to be, why are you implementing it in such a complicated way
>. Secondly, Tilden himself hasn't had a lot of luck in this area --
>I'm not sure if the reason he gives for "Lobster's" failure are correct
>(that is, if it was faced with a complicated situation it just fell down
>and wiggled until things calmed down.) I'm afraid that what really
>happened is that he was going in a direction you seem to be following: the
>diagram for Lobster is nice and regular and pretty and all but no reason is
>given for any of it. Lobster (without much in the way of electronics) is
>the picture used on the BEAM poster & is discussed on pages 11 & 12 of
>"Living Machines" (get a copy from Dave's www.solarbotics.com site) --
>also, the picture at the top of "Living Machines" is the robot.
>
>
>So, once again, I've got to get back to my main point: what is your robot
>supposed to -adapt- to? In other words, what is the point of the exercise
>(I'm not saying that it is a bad idea, btw.)
>
i want to create something that is really adapted to its vary enviroment
something like a spine to all walking robots...i build one microcore and
testes it in a rocky land and sand pile(one of my gears from the servo got
crushed.... Sigh*...there goes 20 bucks..u have any spare servo gears to
spare?)bbut it still aint adaptive as tilden's bigfoot walker cause i saw it
in the beam home made video...so i was thinking if tilden ever used neural
net to stimulate it more...and my robot drags it hind leg a lil...and this
does not happen to Tilden's robot ,i dont know why it is like he used some
sort of electronic that straight away changes its gait when a little motor
stress is felt...i hope u can teach me recreate this behaviour..thanks

and here is a few more thing i would like to know more about lurch the robot
u built....how is doing very adaptive in all enviroment and what gear
motors/electronics is it using?


______________________________________________________

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

eGroups.com home: http://www.egroups.com/group/alt-beam
http://www.egroups.com - Simplifying group communications



Home