Alt-BEAM Archive
Message #00836
To: Richard Piotter richfile@rconnect.com, beam beam@corp.sgi.com
From: Sharon swilliam@cadvision.com
Date: Mon, 22 Feb 1999 22:36:44 -0800
Subject: [alt-beam] Re: 78 Neuron Robot
well, you could have a open string Nv setup, so a single pulse travles
down then stops, completing one cycle, and you could have an Nu or sensor
sending the pulse to the begining of the loop. I am trying to build a
walking bot to compete in the suo compitition at WCRG. I was going to use a
processor to process the info from infa red sensors, but I may not. I was
thinking, I saw this document that used an processor to send an IR pulse,
then check for it with the left sensor, and then again with the right ect.
And I was thinking if you used an open loop, you could have it send a pulse
as it hit the end of the string, and have the detector at the top receive
it and send a pulse down the string. You could use a flip flop or such to
switch between the left and right detector, and if the signal is recieved
by a detector (bounced off by an nearby object) it would send a pulse down
a different string tunned so it turns away from that side. So basicaly, you
would have three strings, forward, turn left, turn right, etc. and have the
IR detectors determine which string to pass a pulse down.
and, if you wanted to get real difficult you could have a backup string so
if both detectors are triggered it backs up. or even more difficult, you
could use Nu's to detect wether or not an object is moving by detecting
whether a detector detects an object then all of a sudden it doesn't and
the other does. Then you could have it turn towards and follow the moving
object, cool, no?
Ok, I will stop my incomprehensible rambling now =^)
-Jeremy Williams
PS. I wouldn't count on my walking sumo bot being finished for the WRCG,
maybe next year. ( will still be there though, hope to see ya there eh!)
At 10:05 PM 2/22/99 -0600, Richard Piotter wrote:
>The original neural net was designed on a computer. A virtual "world" and a
>virtual "bug" were created and it was tested for over 2 years. They proceded
>into building a 1 Kg robot using the same neural network. I don't know if the
>robot was hardware or soft ased, but they mentioned that they had to tune
it for
>it to work, whereas the computer sim didn't.
>
>I'd make the hardware if i could get all the parts. I'd like to go a full
>hardware route to prove that BEAM can do a very complex robot, and
hopefuly go
>for a record! Hehe! (:
>
>I'd not hesitate to go with hardware, the problem is that every schematic
of an
>Nu neuron I've ever seen uses inputs from tactile sensors and has one
output. I
>simply assume the outputs spit to other inputs, but I have yet to see a
diagram
>of two inputs coming together. I also am not certain how one Nu can have both
>inhibitory and excitatory inputs and outputs. I'd ask Tilden... AGAIN...
but I
>never get a reply. You know his address? He seems to have more than one. One,
>little, tiny example and I'd be set! I'd start building this thing in a
>heartbeat! There is no GOOD information on the interconnection of Nu and Nv
>neurons. Also, the subject of multiple Excitatory and inhibitory signals
realy
>interests me. Obviously, biological neurons are going to have this in
quantity,
>but the curent Nu neurons in BEAM are either + or -, presumably excitatory or
>inhibitory, respectively. I may look into op-amps, as I've seen some info
allong
>those lines, but I wonder how complex that'll be???
>
>Who knows, well, maybe Tilden and those scientist guys! Hehe! (:
>--
>
>
>Richard Piotter
>richfile@rconnect.com
>
>The Richfiles Robotics & TI web page:
>http://members.xoom.com/richfiles
>
>
------------------------------------------------------------------------
eGroup home: http://www.eGroups.com/list/alt-beam
Free Web-based e-mail groups by eGroups.com
Home