13435 Re: And now for something completely different...

Alt-BEAM Archive

Message #13435

To: beam@sgiblab.sgi.com
From: Bruce Robinson Bruce_Robinson@telus.net
Date: Fri, 14 Apr 2000 22:05:40 -0700
Subject: Re: And now for something completely different...

Jean auBois wrote:
>
> I've been thinking about what kind of bigger systems could use BEAM
> technology as a basis. One architecture that comes to mind would look like:
>
> BEAM Sensor Array --> Conventional Digital Logic As Neural Net -->
> Latch --> Policy Logic --> Actuators.
>
> One of the important points about the conventional digital logic is
> that we'd be using the same 74... chips that we know and love as a
> sort of neural net. Now, a "normal" neural net depends on > interconnectivity and weights applied to every input that comes into a
> neuron. When you are using digital logic, you've got on or off and
> nothing in between, nothing that could provide weighting. However, I
> have a sneaking suspicion that this wouldn't make any difference
> whatsoever. Instead of levels, what would be important would be
> differences in phase and frequency between the inputs of the elements.

Not too different from what Richard Caudle and I were kicking around a
few months back, Zoz. I tended to lump "digital logic ->latch -> policy
logic" into one fairly complex network of Nv's and Nu's, but you could
certainly divide up the network into those components.

> ... a boolean combination of a set of inputs might produce outputs
> that are recognizable, are useful (a "it is bright to the right of
> me but I've got a touch sensor triggered on that side and I'm low
> on power" kind of result.)

Instead of discrete logic, use an Nu neuron to accumulate impulses from
the different sensors. The impulses could be positive and negative
(inhibitors), and could be weighted differently. The weightings could
even be altered by other elements.

> Now, at this point, there would be a simple latch. The latch would be
> triggered by the overall clock for the device -- its heartbeat if you
> will. As a result, the robot would be taking a snapshot of "recognized
> situations" and would extend it, stretch it for a while. The clock might
> be fairly fast...

I also found the clock to be the key to making the Nv/Nu idea work. The
impulses I refered to come from conventional Nv neurons, which are
triggered by a "clock" impulse. The sensors control the delay of the Nv
neurons, which affects how quickly the charge accumulates on / drains
from the Nu neuron.

The Nu neuron is, in fact, a sort of latch.

The main difference in our approaches is that the Nv/Nu net doesn't
takes snapshots. Everything is processed simultaneously. This may or may
not be an advantage -- it would certainly require some careful design.

> The next section would be policy logic -- in other words, if the
> robots recognizes a given situation, more-or-less what would you
> want its actuators (motors or wheels or whatever) to do.

Entirely do-able with only NAND gates and inverters. I used this idea
for the BEAM Robotics Soccer project. However, I'm pretty sure it can be
done with Nu neurons as well -- I'm still fooling around with that idea.

> ... Since the latch holds the input to the policy logic steady for
> a while, the robot would get a chance to actually complete an action
> before a new situation comes along.

Which is not always good. The key, I think, is to create a hierarchy of
situations. The ones at the top of the hierarchy always take precedence.
Another interesting notion builds on somthing Bob Shannon mentioned a
while back: have the robot monitor itself.

> I fear that this scheme might fall prey to the failing Tilden
> expressed about computer controlled robots -- what do you do if
> a situation occurs that you don't recognize or don't have a policy
> for (or even worse, your policy screws things up).

For sure. That's why I keep forcing myself away from digital logic (sure
is an easy trap to slip into). The Nv/Nu net idea seems to have quite a
bit of room for flexibility, but my early experiments have really
hammered home the fact that tuning the circuits is a royal pain.

In terms of the 40-hour week, most of us spent about 4 person-years of
full time effort tuning our bodies to walk -- and we had a built-in
control system. I suspect there are a lot of "tuning rules" we need to
develop before we can make larger environment-proof systems.

Bruce

13436 Sat, 15 Apr 2000 16:14:20 +1000 Re: Solar powered pendulumn project beam@sgiblab.sgi.com "Ben Hitchcock" Here's one possible solution:

Mount a permanent magnet on the base of the pendulum, pointing North down.

Use a reed switch underneath dead center, mounted so that there is a range
of at least two centimeters where the reed switch is activated. Also mount
the coil below dead center, slightly offset if you feel like it.

Connect the solar panel to a capacitor permanently.

Now use the reed switch to interrupt the power supply to a simple
oscillator, the period of which is equal to a tiny bit less than half the
period of the pendulum at low amplitude. This circuit should turn on the
coil when it first receives power, then turn off after roughly 0.1 seconds
or so. Then it should wait for a that certain time (1/2 T plus a squidgen),
then pulse the coil again.

Now, here's how it works:

Say we have a pendulum swinging away, and light on the solar panel so the
cap is charged. When the pendulum gets near the bottom of the travel, the
magnet activates the reed switch, which turns on the oscillator circuit.
The coil turns on immediately and the pendulum is attracted closer to the
bottom dead center. But the coil turns off quickly so the pendulum is free
to continue back up to the other end of its travel unimpeded. As the magnet
passes away from the center then reed switch turns off, so the circuit turns
off as well, ready to be turned on when the pendulum gets back down from the
other side.

So why the oscillator, I hear you ask?

Well, consider the case when the pendulum isn't moving. Then, the reed
switch is activated all the time, so a quick pulse every now and again will
start the pendulum swinging. While the pendulum is within the reed switch
area, the oscillator will slightly 'beat' the pendulum back to dead center,
making the amplitude increase until the pendulum is swinging nicely.

hope this helps,

Ben Hitchcock
----------
>From: Jeffrey D Spears
>To: beam@sgiblab.sgi.com
>Subject: Solar powered pendulumn project
>Date: Sat, 15 Apr 2000 11:16
>

> Greetings;
>
> A few weeks ago I stumbled onto BEAM and have become interested in
> the whole thing. Seeing how I am taking the second-term circuit
> analysis course (FOCs, SOCs, diodes, mosfets--all both time and freq
> domain)--Beam technology provides excellent test-bed to hammer it
> all home. Fooling around at home, siren RC circuits are boring.
> Beam circuits/applications look cool!
>
> So I have in mind to build a Beam style solar powered pendulumn to
> hang in my window. What I imagine is some sort of simple linear
> motor--a magnet and big coil--delivering a little kick at the right
> time. Once the appratus is built I can experiment with different
> triggering methods, be it detecting the position of the pendulumn or
> a resonate circuit tuned to the period of the pendulumn.
>
> The motor itself may consist of a magnet attached to the pendulumn
> via an arc of stiff metal rod. This arrangement would pass through
> the coil, mounted to the base, as the pendulumn swings.
>
> Intuition tells me that the push wants to be applied during the
> time when the pendulum starts to move down from either side.
> Playing with my pencil and a paper, I cannot see any easy way to
> configure it this way. Therefore my idea is to have the coil in
> the center of pendulumn travel. It will be pulsed with one
> polarity when the pendulumn is traveling one way, and the other
> polarity when the pendulumn is moving the other way. The pendulumn
> itself will be traveling by this point at varying velocites which
> will probably need compensating for.
>
> If all this works out, then a seperate circuit could be added to
> maintain operation overnight.
>
> The PM-1 from Mike Tilden--a super-cap and timing network
> contraption--looks like a fun little circuit and may be suited for
> this application.
>
> Any thoughts from the group? What would N. Tesla do?
> What say you?
>
> ok..jef
>
>
> Jeffrey D. Spears
> University of Michigan
> College of Engineering
>
> ``Double-E, can't spell gEEk without it!''
> -Captain Gerald M. Bloomfield II, USMC
> (my brother)

13437 Sat, 15 Apr 2000 00:30:13 -0600 I hope a 28KB GIF file isn't too big... beam@sgiblab.sgi.com Jean auBois --=====================_15397231==_

which is a more detailed drawing of what I was going on about.

Also a reply to Bruce...

>Instead of discrete logic, use an Nu neuron to accumulate impulses from
>the different sensors. The impulses could be positive and negative
>(inhibitors), and could be weighted differently. The weightings could
>even be altered by other elements.

I feel fairly uncomfortable about this. Back in college when I was
"programming" a Hitachi analog computer, even the best amplifiers and
integrators were pretty flaky. I'm afraid that Nu neurons would be really
too unpredictable. On the other hand, there is a lot of room for
experimentation with the so-called "bias point" (between the passive
components and the inverter) that hasn't been done yet.

>The Nu neuron is, in fact, a sort of latch.

Um, actually, the Nv latch is a latch -- it is an Nv neuron with equal
resistances to both Vcc and ground instead of just the single one to ground
as in a normal Nv neuron. It remembers the "direction" of the last
positive or negative going edge. An Nu is a pulse-stretcher and eventually
loses state and sometimes I think you've got to remember something for a
Real Long Time.

The equal resistances of the Nv latch "eventually" cause the bias point to
go to Vcc/2 which is definitely somewhere in the middle of the hysteresis
curve (providing you are using something like a 74xxx14 chip). A positive
pulse takes the bias point above the upper limit and the output goes to
zero, but the bias point never goes down below the lower limit (likewise
for negative pulses except the output goes to one).

>The main difference in our approaches is that the Nv/Nu net doesn't
>takes snapshots. Everything is processed simultaneously. This may or may
>not be an advantage -- it would certainly require some careful design.

I dunno. I remember agreeing with you (eventually) that beings with
nervous systems are doing a massive amount of -asynchronous-
processing. The best artillery aiming systems designed during/after the
second world war were purely continuous feedback devices and they worked
extremely well. However, the kind of actuators we are dealing with (little
motors, solenoids, fairly small power supplies and the like) have an
enormous amount of hysteresis compared to biological systems and do need
some time to actually get something done.

>Which is not always good. The key, I think, is to create a hierarchy of
>situations. The ones at the top of the hierarchy always take precedence.
>Another interesting notion builds on somthing Bob Shannon mentioned a
>while back: have the robot monitor itself.

Basically that is what subsumption is all about. On the other hand, I've
not seen any subsumptive systems designed with more than three levels of
hierarchy. Although I may be completely wrong, I get the impression that
this sort of design is kind of tough to do even though we people do it all
the time ourselves. I haven't seen any new papers out of the MIT bunch
about subsumption for years now.

Like: I want to drive to the store which is composed of knowing how to
navigate and follow traffic rules and so forth which is composed of
steering, braking, accelerating -- and I've learned the last so well that I
don't even NOTICE them much.

13438 Sat, 15 Apr 2000 01:36:35 -0500 Quadrapod & More BEAM , Twin Cities Robotics Group Richard Piotter HOLY COWS!!!

I promised I was going to update my page on Friday...

Well, It turns out that I kept adding more and more info, that... well,
quite frankly, I don't know what to do with all the info.

I'll have to work on it piece by piece. So far, it's all been images and
graphics. I have yet to touch a character of html. I still have
captions, pointers for images, and some schematics yet to do.

Since there is no way on earth I can do this any earlier than Sunday,
I'll give a list of what I have done and will be in store when the
update is completed and goes online...

1: Basic blueprints for building your own Quadrapod or Hexapod!
Fairly generic design, can use ball or screw ended levers.

2: New images of my Quadrapod, featuring battery/charge system.
Images from multiple angles, plus leg views.

3: Images of the portable A/V display and video components.
4" color LCD, Bass reflex speakers, camera

4: Latest image of Walkman Jr. Frame.
Perfect 37x66 mm size for the new Solarbotics panel.

5: Picture of my _still_ unused BASIC Stamp IIsx. Hehe! :)

6: A few "Misc" items:
Repaired SNES, Homemade Outlet, 1 rev per 8 DAY gearbox!

Obviously, a good majority of it will require editing and lots of text
to describe it. I also used my camera to do the images of the
blueprints. I will eventualy scan them if I have any common sense. If I
scan them, they'll be actual size (72 dpi) and easier to follow. Also,
note that these plans include a thin (1/16, more or less) border around
the cut, so all outside cuts are for the OUTSIDE edge of the blade to
follow, and all inside cuts are the same too. I'll explain it on my web
page when I can finish all the info.

Hope the teaser was worth it. I'll try to hurry. I was going to have
this whole weekend freed up, but as usual, I have things to do at the
worst possible times!

And my stupis BSIIsx still eludes a successful communication via a
school PC, so I'm getting fed up with it. It has an LCD and a "Board of
Education", which is some sort of expirimentation board. Geez! Seems you
need an education just to interface to that little thing!

Opinions and questions are welcome, even if the content isn't completed
and posted yet.
--

Richard Piotter The Richfiles Robotics & TI web page:
richfile@rconnect.com http://richfiles.calc.org

-- Make Money by Simply Surfing the Net or responding to E-Mail!!!
-- Click below!!!

http://www.alladvantage.com/go.asp?refid=ATL147
http://www.spedia.net/cgi-bin/dir/tz.cgi?run=show_svc&fl=8&vid=329630

13439 Sat, 15 Apr 2000 01:47:52 -0500 Re: And now for something completely different... beam@sgiblab.sgi.com Richard Piotter Sounds like a very great idea! The closest I came to this type of system
is my little logic circuit for Quadrapod. Mostly AND, OR, and inverters,
with an XOR or two thrown in. It takes 5 "command" inputs that redirect
the signals from a sequence generator (such as a 6 Nv microcore, or a
counter). Signals can be combined for different functions. Some
functions "replace" the existing commands. Others run new behaviors by
"adding" more than one input.

Networks that do more are a big boost to BEAM. When we can make them
_do_ more, then maybe more people will notice!

Bruce Robinson wrote:
>
> Jean auBois wrote:
> >
> > I've been thinking about what kind of bigger systems could use BEAM
> > technology as a basis. One architecture that comes to mind would look like:
> >
> > BEAM Sensor Array --> Conventional Digital Logic As Neural Net -->
> > Latch --> Policy Logic --> Actuators.
> >
> > One of the important points about the conventional digital logic is
> > that we'd be using the same 74... chips that we know and love as a
> > sort of neural net. Now, a "normal" neural net depends on > interconnectivity and weights applied to every input that comes into a
> > neuron. When you are using digital logic, you've got on or off and
> > nothing in between, nothing that could provide weighting. However, I
> > have a sneaking suspicion that this wouldn't make any difference
> > whatsoever. Instead of levels, what would be important would be
> > differences in phase and frequency between the inputs of the elements.
>
> Not too different from what Richard Caudle and I were kicking around a
> few months back, Zoz. I tended to lump "digital logic ->latch -> policy
> logic" into one fairly complex network of Nv's and Nu's, but you could
> certainly divide up the network into those components.
>
> > ... a boolean combination of a set of inputs might produce outputs
> > that are recognizable, are useful (a "it is bright to the right of
> > me but I've got a touch sensor triggered on that side and I'm low
> > on power" kind of result.)
>
> Instead of discrete logic, use an Nu neuron to accumulate impulses from
> the different sensors. The impulses could be positive and negative
> (inhibitors), and could be weighted differently. The weightings could
> even be altered by other elements.
>
> > Now, at this point, there would be a simple latch. The latch would be
> > triggered by the overall clock for the device -- its heartbeat if you
> > will. As a result, the robot would be taking a snapshot of "recognized
> > situations" and would extend it, stretch it for a while. The clock might
> > be fairly fast...
>
> I also found the clock to be the key to making the Nv/Nu idea work. The
> impulses I refered to come from conventional Nv neurons, which are
> triggered by a "clock" impulse. The sensors control the delay of the Nv
> neurons, which affects how quickly the charge accumulates on / drains
> from the Nu neuron.
>
> The Nu neuron is, in fact, a sort of latch.
>
> The main difference in our approaches is that the Nv/Nu net doesn't
> takes snapshots. Everything is processed simultaneously. This may or may
> not be an advantage -- it would certainly require some careful design.
>
> > The next section would be policy logic -- in other words, if the
> > robots recognizes a given situation, more-or-less what would you
> > want its actuators (motors or wheels or whatever) to do.
>
> Entirely do-able with only NAND gates and inverters. I used this idea
> for the BEAM Robotics Soccer project. However, I'm pretty sure it can be
> done with Nu neurons as well -- I'm still fooling around with that idea.
>
> > ... Since the latch holds the input to the policy logic steady for
> > a while, the robot would get a chance to actually complete an action
> > before a new situation comes along.
>
> Which is not always good. The key, I think, is to create a hierarchy of
> situations. The ones at the top of the hierarchy always take precedence.
> Another interesting notion builds on somthing Bob Shannon mentioned a
> while back: have the robot monitor itself.
>
> > I fear that this scheme might fall prey to the failing Tilden
> > expressed about computer controlled robots -- what do you do if
> > a situation occurs that you don't recognize or don't have a policy
> > for (or even worse, your policy screws things up).
>
> For sure. That's why I keep forcing myself away from digital logic (sure
> is an easy trap to slip into). The Nv/Nu net idea seems to have quite a
> bit of room for flexibility, but my early experiments have really
> hammered home the fact that tuning the circuits is a royal pain.
>
> In terms of the 40-hour week, most of us spent about 4 person-years of
> full time effort tuning our bodies to walk -- and we had a built-in
> control system. I suspect there are a lot of "tuning rules" we need to
> develop before we can make larger environment-proof systems.
>
> Bruce

--

Richard Piotter The Richfiles Robotics & TI web page:
richfile@rconnect.com http://richfiles.calc.org

-- Make Money by Simply Surfing the Net or responding to E-Mail!!!
-- Click below!!!

http://www.alladvantage.com/go.asp?refid=ATL147
http://www.spedia.net/cgi-bin/dir/tz.cgi?run=show_svc&fl=8&vid=329630

13440 Sat, 15 Apr 2000 00:57:58 -0600 The simplest part of the idea might be the most difficult part to understand beam@sgiblab.sgi.com Jean auBois The BEAMAnt unicore circuit is composed of a bicore "embedded" in another
one. Each of the bicores have a direct electrical effect on each
other. Because of this, the result is pretty complex and rather hard to
characterize.

However, we could consider two entirely separate bicores with no
appreciable electrical interaction. Each one of those can be used as a PWM
generator, with the pulse width depending on whatever sensor you were
interested in using.

Let's say that somehow you've tuned each bicore so that their basic
frequency was pretty close. Furthermore, let's say that you take the
output from each of them and run it through separate inputs of an AND
gate. The result would be a boolean product of the two individual
signals. If the duty cycle happened to be 50% then the output would depend
on the delta phase angle between the two signals -- it could range anywhere
from 0 to 50% (180 degrees out of phase to 0 degrees out of phase.)

In spite of the fact that we are using a purely boolean device -- the AND
gate -- we can obtain a result that has a continuous and that has a very
smooth gradation between the two extremes. We don't need to resort to an
infinitude of voltages.

The other case, which I won't go into here right now, concerns what happens
when the two signals are of different frequencies. There you'd get some
version of a beat frequency sort of signal which boggles the mind in its
complexity but probably could be characterized with delta frequency on one
axis and starting delta phase angle on the other.

In other words, digital devices producing extremely analog effects.

Anyhow, it's late, gotta go to bed and all...

jab

Home