home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!munnari.oz.au!metro!extro.ucc.su.OZ.AU!andy
- From: andy@extro.ucc.su.OZ.AU (Andrew Miehs)
- Subject: Beginners Question: Walking Robots & NNs
- Message-ID: <1992Dec12.042020.6232@ucc.su.OZ.AU>
- Sender: news@ucc.su.OZ.AU
- Nntp-Posting-Host: extro.ucc.su.oz.au
- Organization: Sydney University Computing Service, Sydney, NSW, Australia
- Date: Sat, 12 Dec 1992 04:20:20 GMT
- Lines: 28
-
- I've read many articles recently (in SciAm and others) about insect-like robots
- which teach themselves how to walk, from scratch.
-
- The articles didn't go very far into the theory of it, but it seems, to my
- limited knowledge, that this would be an appropriate application for a neural
- net. I'm have recently been teaching myself about NNs and this problem
- intrigued me such that I'd love to try it out myself.
-
- Does anyone know much about these 'creatures'? I am especially curious as to
- how they know whether their actions are "successful" and even how they can
- define what a "succefsful action" is.
-
- I was thinking of working on a much simpler scale - a simple robot which had
- four independently motored wheels, whose on/off and forward/reverse motion
- could be controlled by a computer. It's objective, would be to find some
- transmitter, given intensity signals from four directional receivers (probably
- microphones & ultra-sonic sound).
-
- Am I being hopelessly naive in believing that I could create a NN to solve
- such a problem? If not (i hope not) what approach would you recommend?
-
- Malcolm Ryan (care of Andrew Miehs)
- Email replies to andy@extro.ucc.su.oz.au
- --
- +-------------------------------+---------------------------------------------+
- | Andrew Miehs | |
- | Mail: andy@extro.ucc.su.oz.au | $$$ THIS SPACE FOR RENT $$$ |
- | NB: Don't call me andy! | |
-