DPRG List  

DPRG: Neural Network(Yes!) robot revised

Subject: DPRG: Neural Network(Yes!) robot revised
From: Alan Bredon alanb at cryogen.com
Date: Wed Jan 14 09:08:53 CST 1998

Corey Hansen wrote:[..snip..]

> The robot then goes to the address in
> the code which corresponds to the sensor inputs. Now the robot needs
> to
> decide on the motor output remembering the values of the 7 variables
> representing the outputs. Now that the robot has the output figured
> out it
> starts it. Now it waits until a sensor input(s) change(s). Now it
> compares
> the older input with the new. If the input is lower then before then
> the
> robot upgrades the  previously used output degrades the others  , or
> vice
> versa if the input is higher. Then the robot repeats this , learning
> the
> best use of outputs of the most common inputs.


Now this is true AI!  I read about this technique in a book on robotic
AI back in the early 80's.  I think I will try to find it and maybe
excerpt some of it here, soon.

The inputs are addresses and the outputs (motors) are data.  Different
combinations of bumper strikes address different locations in memory
which contain different motor drive configurations.  You can add bits to
represent and trigger more complex stored command sequences and bits to
keep track of the popularity of a particular behavior attempt for
particular input combination.  It is also good to go through all memory
locations and lower the popularity bits over time and to add a certain
amount of randomness to the system to prevent repetitive action,
although some behaviors may end up repetitive anyhow.

I'll put more info up after I review the book.

- --
     _  _
    / || \    Regards, Alan Bredon
   /  ||  \   My views & opinions do not represent my company.
  /. .||. .\  My email address is MINE.  Spammers should pay me!
 /-|-||||-|-\ Legit replies delete the -NOSPAM.
  _|_|__|_|_  http://www.imagin.net/~abredon
 | . .  . . | And always remember to ESCHEW OBFUSCATION!


More information about the DPRG mailing list