Alt-BEAM Archive

Message #04806



To: "'beam@sgiblab.sgi.com'" beam@sgiblab.sgi.com
From: Wilf Rigter Wilf.Rigter@powertech.bc.ca
Date: Tue, 22 Jun 1999 22:18:54 -0700
Subject: [alt-beam] Re: Ah! Blessed Skepticism! (was: Feedback explained?)



content-type: text/plain;

I was impressed by many of the previous comments on this thread (not
including my own).

Most of Mark T. comments were on microcores and >4Nv cores and some are very
puzzling indeed. Take the following statement:


What is surprising
though is that a 4 neuron microcore with a single process in it can
still be influenced by it's prior passing over a neuron 3 neurons
ago. It can "see" itself and be repulsed.

Try it. Dead short a cap without a process on it, remove quickly
and watch the process rate change in a microcore. Processes live
in a tenuous voltage space very susceptible to any charge
influence.

Neuron outputs trigger on the log decay of the coupling cap before
the inverter input, which is a very shallow voltage curve and
subject to any charge influences prior to it. Exact equations are
pending, but as many of you know, a 5% variation in coupling
capacitors can mean seconds in variation characteristics.

But you didn't hear it from me, cause you won't believe it until
you can use it.

markt.

This statement about a "memory effect" of previous processes in Nv cores
must have been based on some theoretical work since these effects are
probably caused by an average dc component of the process waveform shifting
the trigger level which shows up as a term in equations that do not include
the "dc restoration" effect of the commonly used HC logic family input
protection diodes. However if this "memory effect" is desired or considered
useful then the simple addition of resistors in series with the Nv bias
point can re-introduce this "DC component term".

<>
Now I agree that garbage in equals garbage out no matter how much positive
or negative feedback you put in the loop and all the mumbo jumbo aside, I
did this bench test:

To a standard suspended Bicore add a BG micro motor directly to the Bicore
outputs. Observe the rotation and Bicore pulsewidth is about 90 degree each
way and 1 second pulsewidth duration. Now load the output shaft with some
finger pressure (not a complete stall) and note the rotation is about 30
degree each way and the pulse width is 0.5 seconds. So there it is
mechanical AND electronic feedback. Obviously the rotation is reduced by
mechanical loading BUT the pulsewidth is also reduced. This appears to be
caused by a combination of motor noise coupled through the Bicore timing
caps pre-emptively triggering the Bicore and/or the effect of reduced Bicore
output voltage swing coupled through the timing capacitors affecting the
Bicore time constant.

It seems to me that in a walker this MAY be a useful feedback by reducing
wasted power in case of a stuck leg and it MAY be safer to change walking
gait to a shorter shuffle on slope to avoid tipping over or on a high
resistance surface it MAY produce a more efficient gait.

If the Bicore output is buffered before connecting to the motor, the output
pulse duration is actually slightly longer when the motor is loaded even if
a small resistor is introduced in series with the powersupply to reduce
supply voltage under load (so much for powersupply feedback).
Next I will add some gain to this motor current feedback signal to see if
these effects can be more precisely controlled.

enjoy (although that's starting to sound like the Prisoner's "be seeing
you")

Wilf Rigter mailto:wilf.rigter@powertech.bc.ca
tel: (604)590-7493
fax: (604)590-3411

> -----Original Message-----
> From: Bruce Robinson [SMTP:Bruce_Robinson@bc.sympatico.ca]
> Sent: Tuesday, June 22, 1999 6:35 PM
> To: beam@sgiblab.sgi.com
> Subject: Re: Ah! Blessed Skepticism! (was: Re: Feedback explained?)
>
> James Wilson wrote:
> >
> > Here's my cents worth!
> >
> > When a motor is stalled (leg stuck) the inductor draws as much current
> > as possible from the power source. The motor then happens to move if
> > even a little bit and this will cause a back emf to increase the
> > resistance of the circuit and thus influence the nervous net. Is this
> > not apparent?
>
> and John A. deVries II wrote:
>
> > Ok, sure, people have -some- sort of qualitative observations indicating
> > that a nervous net robot acts differently when a leg is stuck than when
> > they the legs are free, but NO ONE seems to have made any repeatable
> > experiments that either support or deny any particular hypothesis.
>
> So let's see. Ian brought up the feedback idea. I refuted it. Ben
> pointed out a flaw in my argument. I mentioned that there COULD be some
> feedback at certain points in the walking cycle IF the supply voltage
> was sucked way down. Steve pointed out that we were going over old turf.
> And John said no one's actually done a proper experiment to
> prove/disprove it.
>
> After thinking about it some more, I realized that the main cause of
> feedback to the Nv net would be when a motor was really stalled -- and
> in that case the mechanical effects (motor barely moves) greatly
> outweigh the Nv effects (Nv delay is shortened, reducing the amount the
> motor will move).
>
> And the previous discussions that Steve so kindly sent us (thanks,
> Steve!) seem to support this point of view. When I look at the walker as
> a complete system, I return to my original assertion, that inherent
> feedback to the Nv is bogus. When (if) it IS present, the
> electro-mechanical conditions that cause it have a far greater effect.
>
> So to go back to John's statement, show me a repeatable experiment that
> clearly demonstrates the effect, and I'll agree it's significant.
> Otherwise I'm a confirmed skeptic.
>
> And I promise not to say any more about this for a long, long time .
>
> Take care,
> Bruce


------------------------------------------------------------------------

eGroups.com home: http://www.egroups.com/group/alt-beam
http://www.egroups.com - Simplifying group communications





Attachment: Nvmemory.gif

Home