To LUGNET HomepageTo LUGNET News HomepageTo LUGNET Guide Homepage
 Help on Searching
 
Post new message to lugnet.roboticsOpen lugnet.robotics in your NNTP NewsreaderTo LUGNET News Traffic PageSign In (Members)
 Robotics / 13141
13140  |  13142
Subject: 
Re: AI in RCX
Newsgroups: 
lugnet.robotics
Date: 
Wed, 15 Nov 2000 14:53:01 GMT
Original-From: 
Yiu <H9819152@HKUSUA.HKU.stopspamHK>
Viewed: 
796 times
  
In fact, I had already worked out the 1st NN
by working on the RCX with my PC using VB
(after studying NN hardly as I am fresh to it)

This NN is in the sturcture of Hamming network that can recognize the
environment prototype by reading the Ultra-sonic distance sensor (Thanks
to
JB's cool sensors)
Right corner, T-cross etc.
More advance structure is not chosed as I have to consider the
limitation of
RCX

Before moving on to next NN that teach the RCX to follow a wall,
I want to integrate the 1st NN into the RCX first and test its
performance
in the autonmous mode.
(Seem to be the most difficult part)

However, I am still struggling on the legOS.
Considering how to use the resources more efficiently and studying the
math
function.

Hope I could work out one more little step soon.
(I have a lot of assignments to do currently @_@)

Yiu


Tomasz Salamon wrote:

I hope that the subject of AI in RCX is not discarded yet.

The project is really challenging, but I think worth to be taken.

Controlling robot is a complex task, or rather a set of tasks. I think

that it would be a good (practical) approach to separate simple tasks
from more complex task (or task which require AI and create so called
hierarchical controller.

Simple tasks I mean controlling i.e. speed of motion (which can be
easily and sufficiently done by deterministic {not NN!} algorithm) and

higher level task like decision concerning the direction of movement.

In that configuration you can significantly reduce the NN size (less
layers, less neurons) => RCX computational power may be sufficient.

I don't know what interface to NN you use. I think that Matlab with NN

toolbox (from The MathWorks) can be useful to first simulation, • learning
and testing of NN on PC. You can even estimate the computation power
needed for NN.
You can also generate a C code for the NN by means of Matlab and RTW • and
after some enhancement you can probably place the code to your RCX.

Good luck!
Keep us informed about your progress.

tomasz salamon



1 Message in This Thread:

Entire Thread on One Page:
Nested:  All | Brief | Compact | Dots
Linear:  All | Brief | Compact
    

Custom Search

©2005 LUGNET. All rights reserved. - hosted by steinbruch.info GbR