Alt-BEAM Archive
Message #09293
To: beam@sgiblab.sgi.com
From: Bruce Robinson Bruce_Robinson@telus.net
Date: Mon, 17 Jan 2000 16:22:32 -0800
Subject: [alt-beam] Re: Interseting, additional idea..........................
Richard Piotter wrote:
>
> If you look at my web page and look under Quadrapod, you'll find a link
> to a schematic. It uses an Nv loop to generate a drive sequence, but
> gates to translate 5 input signals (Stop, Forward, Reverse, Left, and
> Right) into changing pathways for the Nv processes, resulting in 9 types
> of gaits (Stoped, Forward, Reverse, Rotate Left, Rotate Right, Walk
> Forward & Left, Walk Forward & Right, Walk Reverse & Left, and Walk
> Reverse & Right).
AND
Meabadboy@aol.com wrote:
>
> As for the BRAIN or central processor of the creature ~ Has anyone
> looked at the possibility of useing AND/NAND gates as a make up for
> decision logic along with the RC timeing network of the inverter?
I remember trying to analyze Richard's Quadrapod circuit when I first
got into BEAM. Never did entirely figure it out, even though I was
familiar with logic circuitry.
Since then I've built a number of complex nervous nets and wrestled with
the problem of making them extremely flexible. Here's an entirely
different scheme.
Build a separate nervous net for each gait or motion you want (that is,
a separate circuit for each different arrangement of Nv's). For each
circuit provide two inputs ... a "turn off" input (kills all processes,
using diodes), and a "turn on" input (starts up precisely the processes
you want, using diodes and a couple of inverters).
To control the gait, you kill the circuits you don't want, and start the
circuit you do want.
To operate the motors/drivers, you need to combine the outputs from all
the gait circuits. This is easily done with NAND gates, which happen to
be available in a great many configurations: 2-input, 3-input, 4-input,
8-input.
I've done this on the breadboard, and it works very nicely. The big
advantage is you can tune each neural net circuit separately, so each
one works exactly the way you want for a particular gait. If you build
each neural net on a separate circuit board, you can add or replace
them to experiment with new gaits.
I can't give you a precise comparison to the method Richard used until I
figure out all the possible combinations of his inputs. My rough
estimate is that the chip count would be the same, assuming each of his
input combinations needed its own neural net.
Not necessarily a better idea, but certainly a different approach.
Details on request.
Regards,
Bruce
9294 Mon, 17 Jan 2000 19:02:06 EST [alt-beam] Re: D1 beam@sgiblab.sgi.com Bumper314@aol.com When I tell the list about these problems I am sure to try just about
everything in my power first to make sure im not doing something stupid, so I
have tried several LEDs, I actually haven't even tried the blue one yet.
Tried Several different capacitor values. But still nothing...The LED shines
all the time, one the voltage is enough forward bias the LED. I just dont
know whats up. I hooked an o-scope up to the caps and they arent
discharging...Any ideas
Steve
9295 Mon, 17 Jan 2000 19:30:27 -0500 [alt-beam] Re: D1 beam@sgiblab.sgi.com Bob Shannon Wilf Rigter wrote:
> Hey Bob,
>
> What's a smart cap circuit?
>
> regards
>
> wilf
The Smart Cap circuit is my solution to the D1 dilemma, I developed it for
Chiu's first photovore contest to add long term behavior abilities to the
basic
photopopper design.
I wish I had an electronic schematic, but...
All this circuit uses is a NPN phototransistor, a PNP switch, and a single
resistor.
The NPN phototransistor is connected across the base and emitter of the
PNP switch transistor, with the phototransistors collector going to the
PNP's emitter. This connection is also the positive supply line.
The emitter of the NPN phototransistor is connected to the base of the
PNP, and
also to a bias resistor to ground. For the common Radio Shack NPN
phototransistor (2 lead type, looks like an LED) a value of 5.1 K seems to
work
well. Adjusting this bias resistor sets turn-on point, and adding a cap
can control
any hysteresis if desired (you can also play with the base lead of 3
terminal phototransistors to get the same effect).
The load connects between the PNP transistors collector and ground.
Any load will do, so long as its within the PNP transistors abilities.
The only leakage current (very low) flows while light (and energy) is
available. I find this seems to work better than circuits with a dark
leakage
current (low-side NPN switch with cds photocell).
In Vore-n-more, I used this circuit to dump a 1.5 F gold cap through 75
ohms
and into the capacitor bank of a photopopper. The 1.5 F cap was charged
through a schottkey diode from the photovores 3733 solar cell, through a
small set of 'floater' solar cells in series.
The idea here is that the main solar cell will charge the 1.5 F cap to a
bit
below the photopopper's turn on point, but that does the bot no good when
the
lights go out. I used 1381-J triggers, so I needed the 1.5F cap to charge
to around
3 volts or so to have a useful energy reserve. A pair of tiny AM-1437
cells do this trick nicely.
Another feature of this setup is that an empty 1.5 F cap will put the
photovore to 'sleep' under a bright light, until its fed long enough to
have a good energy reserve.
As the supercap charges, the photovore slowly wakes and turns to the
light, then becomes as active as usual once its full.
'Till the local light pool goes dark....
Then the smart cap circuit triggers, and the PNP switch is no longer
starved for base current. The supercap is dumped through 75 ohms into the
photovores main
drive caps, and the bugger takes off like a cockaroach directly towards
the next nearest light source (there's an on-line vid of this behavior).
Just adding the 1.5 F cap, floater cells, a diode and resistor, the range
of behavior has really expanded a lot. This little photovore has long
term behaviors and 'memory'.
Its really amazing to see how well the photopoppers 'vision' system can
operate over a huge range of illumination levels and still accurately
track a distant light source. I mostly build CPU based bots, and getting
the optical dynamic range that the simple popper has is not easy to do.
But put a little energy reserve in there, and see how well a photovore can
see in the dark!
Home