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 / 7368
7367  |  7369
Subject: 
RE: Mindstorms/reinforcement learning
Newsgroups: 
lugnet.robotics
Date: 
Wed, 13 Oct 1999 02:59:45 GMT
Viewed: 
654 times
  
Please let the rest of us know as well, Ralph- I'm very interested in any
type of AI stuff on the Mindstorms. As a "quick" note, our projects (using
something called Q learning if anyone out there knows the terminology)
involve doing roughly 6*6*7*8*50 calculations in each learning loop. This
can take a substantial amount of time (3-5 seconds) on the bot, and I
really don't think that too much can be optimized, unfortunately.
Furthermore, because the calculation involves a matrix of 6*6*7 doubles,
transferring it to a PC repeatedly would probably take as long, if not
longer. If you take *that* off of the RCX, you might as well just run the
whole thing in simulation. oh well... it is still really, really cool
when you have realized that your bot has learned something, even if it
has happened in painfully slow 3-5 second intervals. More details later...
-Luis


On Wed, 13 Oct 1999, Ralph Hempel wrote:

P.P.S. NQC is great, but you need a large matrix for RL, and for that
legOS is your only choice. Take a look at www.noga.de/legOS/ for more
info on that. Good luck...

My prompt and humble apologies for forgetting Forth, yet again :) That
would also give you the necessary capabilities, though Ralph would have
to speak to it's speed when doing (literally) tens of thousands of
mathematical operations.

Ha, and it wasn't even me that reminded you this time...

Seriously, I am exchanging emails with someone who is doing an AI project
using pbFORTH. He claims the thinking time is a bit slow, but I don't
know enough about the project to make a guess as to why.

I'm wating for his permission to forward the discussion to the pbFORTH
group which has been ominously quiet for a few months now...

Cheers,

Ralph Hempel - P.Eng

--------------------------------------------------------
Check out pbFORTH for LEGO Mindstorms at:
<http://www.hempeldesigngroup.com/lego/pbFORTH>
--------------------------------------------------------
Reply to:      rhempel at bmts dot com
--------------------------------------------------------



#######################################################################

     Profanity is the one language that all programmers understand.
                  -Anonymous

#######################################################################



Message has 1 Reply:
  RE: Mindstorms/reinforcement learning
 
(...) Declaring variables (almost any number of them) in pbFORTH is like this: VARIABLE FOO Fetching and storing is like this: 23 FOO ! ( this stores the value ) FOO @ ( this fetches the value ) Values are 16 bit integers - 32 bit doubles use a (...) (25 years ago, 13-Oct-99, to lugnet.robotics, lugnet.robotics.rcx.pbforth)

Message is in Reply To:
  RE: Mindstorms/reinforcement learning
 
(...) Ha, and it wasn't even me that reminded you this time... Seriously, I am exchanging emails with someone who is doing an AI project using pbFORTH. He claims the thinking time is a bit slow, but I don't know enough about the project to make a (...) (25 years ago, 13-Oct-99, to lugnet.robotics)

9 Messages in This Thread:


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

This Message and its Replies on One Page:
Nested:  All | Brief | Compact | Dots
Linear:  All | Brief | Compact
    

Custom Search

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