Alt-BEAM Archive

Message #13446



To: beam@sgiblab.sgi.com
From: Bruce Robinson Bruce_Robinson@telus.net
Date: Sat, 15 Apr 2000 12:05:42 -0700
Subject: Re: A reply to Zoz ... and a challenge for advanced/bored BEAMers.


Hi, all.

The following discussion is relevant to the challenge ... but if you
find the talk boring, skip to the end of this note, where I've spelled
out the challenge.

Jean auBois wrote:
>
> Also a reply to Bruce...
>
>> Instead of discrete logic, use an Nu neuron to accumulate impulses from
>> the different sensors...
>
> 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. ...

Well they certainly aren't as predictable as digital devices :)

My first tentative foray into using Nu/Nv networks in place of digital
devices was the "beep" counter I devised for Richard Caudle's HPV
project (subsequently improved by Wilf). The prototype would count ten
impulses and then send an output signal and reset itself. It proved to
be very reliable, with an accuracy of +/- 10%. In other words, it would
usually count to 10, but sometimes only to 9 and sometimes to 11.

A more ambitious undertaking is a circuit to estimate a wheeled robot's
direction by counting impulses sent from the moving wheels. Basically
one Nu has 4 inputs -- a "count up" and a "count down" signal from each
wheel. The total charge on the Nu capacitor indicates the direction.
This is proving troublesome, but I'm slowly solving the problems.

These kinds of applications are probably not the best place to use
analog devices. On the other hand, accumulating sensor information to
make a decision seems to be a great place to use an Nu neuron with
multiple inputs. One interesting way to make the Nu response more stable
is to have the neuron inhibit its "competitors" slightly after it fires.
This appears to be a technique used in mammals.

I suspect you will also have to keep the number of inputs to a neuron
relatively small. For complex "decisions", you will probably need to use
lower level Nu neurons to summarize information, which is then passed on
to higher level neurons.

>> 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.

Aha! I hadn't thought of that. I just parked that idea in my "toolbox".

> An Nu is a pulse-stretcher and eventually loses state ...

That's one of many applications for it.

I agree that losing state is a problem ... my direction "memory" device
degrades about 5% over ten minutes, but I REALLY have to tweak it to
pull that off.

The question boils down to, "How long do you want to latch a particular
condition?"

> ... I remember agreeing with you (eventually) that beings with
> nervous systems are doing a massive amount of -asynchronous-
> processing...
> ...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.

True, but this is only the "output" level. I think it's important for
"internal" processing to carry on while the output devices are doing
something. This means A) when an output task is finished, the robot can
respond immediately to the current state of affairs, and B) the output
task can be interupted where appropriate.

>> ... The key, I think, is to create a hierarchy of situations.
>> The ones at the top of the hierarchy always take precedence.
>
> 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.

I have only a vague idea of what subsumption involves. I've drawn my
ideas from studies I did in human psychology back in the mid-eighties...
learning and motor skills mostly.

In the BEAM soccer project, I devised a set of hierarchies that allow
for crude soccer-playing. Interestingly, the deepest level is three :)
Have a look at

http://www3.telus.net/rfws/soccer/m_bhv_fp.html

Actually, your observation on 3 levels maximum is interesting. In the
soccer-playing hierarchy, there were typically 5 to 6 levels of decision
making required before the actual outputs are triggered. However, it
made more sense beyond the third level to use a sequential pattern
generator rather than a decision process. This also seems to be how
humans "work".

> 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.

Here's an interesting observation for you. I've tried this approach with
both a team of soccer playing robots, and a "search and destroy"
fire-fighting robot. The soccer 'bots are in a highly interactive,
rapidly changing environment; the fire fighting robot is in a static,
highly constrained environment. Yet it has proven much easier to
generate a set of soccer-playing hierarchies than it has to generate a
set of search hierarchies.

> I haven't seen any new papers out of the MIT bunch about subsumption
> for years now.

I suspect it has limited application. We humans probably use it as just
one of many integrated strategies. That's why I would like to use analog
"neurons" to build hierarchies rather than digital devices.

--------------------------- The Challenge ---------------------------

Here is a challenge for BEAMers that are getting bored with solar
engines and microcores. Try building a robot that tries to avoid getting
stepped on. Zoz's diagram is an excellent model ... do it in
BEAM/digital as he suggests, or try it with 99% BEAM as I advocate.

The avoidance part is easy: Move out of the way, hide, or attract
attention (flash a light, make a noise, wave a flag, dance a jig). These
have all been done by various people.

The sensors are the tricky bit. You probably need to use several sensors
to be "sure" that a human/animal/predator is approaching. Sound,
proximity, vibration, heat detection ... these all come to mind.

To add to the challenge, avoidance activities should be appropriate to
the conditions. Dancing a jig in the dark isn't likely to attract
attention :)

You can also extend the idea to anti-predator activities. Squirm or make
a loud noise when picked up (OK, no scalpels or lethal voltages please
:)

I'd love to hear about your successes (and failures as well), and I'll
be happy to help if you decide to go the all-BEAM route.

Enjoy,
Bruce



13447 Sat, 15 Apr 2000 13:22:17 -0600 Re: Solar powered pendulumn project beam@sgiblab.sgi.com Dave Hrynkiw At 07:16 PM 4/14/2000 , Jeffrey D Spears wrote:
>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.


Take a look at this little patent: 5488131

It's a self-powered sign-waving device brought to our attention by Douglas
Sasse. It's originally a Japanese patent dating back to '92, (from memory).
Take a look at the patent, which is very large and detailed! It'll let you
do what you want to with very little power. I'm really impressed by this
design. It was found by Douglas as a lotto sign-waver at a local
convenience store. Go figure, eh?

It has a HUGE coil, mounted vertically on a PCB with 5 resistors 2
transistors, and two electrolytic caps and 1 mica cap, powered by a tiiiiny
Panasonic 2V 25x5mm calculator cell. The pendulum has a pair of magnets
mounted so they have the coil pass between them every swing. Haven't spent
the time analyzing the patent yet, but from first appearances, is a very
elegant circuit. Something I would expect Wilf or Ben to come up with!

This little pendulum works on obscenely low-light levels. Really an
efficient design.

Regards,
Dave

---------------------------------------------------------------
"Um, no - that's H,R,Y,N,K,I,W. No, not K,I,U,U, K,I,_W_. Yes,
that's right. Yes, I know it looks like "HOCKYRINK." Yup, only
2 vowels. Pronounciation? _SMITH_".
http://www.solarbotics.com



13448 Sat, 15 Apr 2000 15:48:36 EDT 2N7000 MOSFET beam@sgiblab.sgi.com BUDSCOTT@aol.com I just got some 2N7000 MOSFETs and have a question. I've got the three pins,
S, G, and D. How do they corespond to the schematic?
|---------- 2
1 ------|
|<--------- 3

Which is which?

-Spencer

<http://www.botic.com/users/beamstop>

not a robot scientist
not a college major
not a grad student
not a professor
not a very organized person
just Spencer (isn't that impressive enough?)



13449 Sat, 15 Apr 2000 22:19:58 +0100 (BST) Robot Books beam@sgiblab.sgi.com =?iso-8859-1?q?John=20Lane?= Hey everyone, Is there anywhere in the UK that sells
decent, CHEAP, books on robots?

Jay

____________________________________________________________
Do You Yahoo!?
Get your free @yahoo.co.uk address at http://mail.yahoo.co.uk
or your free @yahoo.ie address at http://mail.yahoo.ie



13450 Sat, 15 Apr 2000 22:01:31 GMT another walker question beam@sgiblab.sgi.com "Alex Hegyi" Where can I get some good reliable and cheap motors for a miller walker?
______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com

Home