Alt-BEAM Archive
Message #10460
To: beam@sgiblab.sgi.com
From: Bob Shannon bshannon@tiac.net
Date: Sat, 19 Feb 2000 10:55:51 -0500
Subject: [alt-beam] Re: reconfigurable Robots: Severing the master/slave bicore connection.
Dennison Bertram wrote:
> Here's another one from the archives of my old notebook:
>
> reconfigurable robots:
>
> If you check out the circuity behind the Stryder type robot, or for the
> Master/slave bicore arrangement you'll notice that the bicores are
> influenced by other bicores. In particular, a resistor from one bicores
> output is used as the input for the another bicore. The first thing one
> should be thinkng about when your thinking about modular robots in beam is,
> how to replace that resistor. After all, that's the only connection between
> the two bicores, with out them, the bicores keep doing their own thing.
I've experimented with modular bicores along these lines. From what I
beleive I was on the bench, the best solution is NOT to couple the bicores
via resistors at all. The Bicores sould be coupled ONLY by feedback from
sensors that measure the overall bots behavior.
This is how the remaining machine is best able to adapt to the changing
performance when a module fails.
In fact, I added a microcore to disable one of three bicores, and removed ALL
the coupling resistors between the three bicores. The behavior of the robot
became more rebust, as it was constantly readjusting to a changing set of
controlling
bicores, sensors, and driven motors.
> (NOTE: Mark Tilden comments that one of the advantages of BEAM is it's
> hardyness. Specifically in his nervous net's he comments how much more
> robust they are than circuits. He states that computer robots are 'one snip'
> away from failure. I'd like to point out, that the earlier microcore was
> also. If you broke any connection in that chain, your bot was dead. However
> if we look at Master/slave arrangements, you'll notice that this is no
> longer true. Snip! the connection between the two bicores, and while their
> actions may not be as cordinated as you would like, the bot keeps ticking.
First off, there is no reason processor based robots cannot be made just as
modular as BEAM tech. There is no real advantage here.
If you design the bot with the right kind of feedback paths, you might notice
that the 'coordination' of the robot appears to go down, but it may actually
perform a lot better, and under a wider range of conditions too.
> Of course you could always snip the bicore...) Anyway, The idea behind
> makeing reconfigurable robots is replacing that resistor, or somehow makeing
> it so the resitor isn't a physical connection. How would one go about doing
> this?
This is exactly what I was addressing in the discussion of geometery and sensor
mapping in the text descriptions of LEM. See:
http://www.the-nest.com/bshannon
> Well back in the day I had several idea's which I never tested (I will
> this summer) but they are as follows:
> 1. Electrical Contact plates
> 2. Magnetic Coils
> 3. CDS cells and small lights
> 4. IR LED's and IR Photodiodes
How about letting each module 'feel' the actions of the others. If you place
sensors so they are effected by other modules, or by the collective actions
of all modules I think you will be very close to what your looking for.
Also you can provide higher levels of behavior by controlling which modules
are active at any given time.
This approach also bypasses the problems of scaling BEAM circuits up to higher
levels of complexity, by keeping the feedback in the real world, rather than in
the electronic domain.
> I'm sure there are other ideas, sonar etc... Heat & thermistors, but you get
> the idea. Instead of building a frame you could build small 'units' perhaps
> a containing a battery, gearmotor, leg, and a bicore. Let this unit flail
> around like a turbot in a RJP until it comes across another simmilar robot,
> let magnets connect them together strongly, and suddenly depending on their
> connection on robot's bicore begins to influence the other's bicore.
> Suddenly the two peices begin to operate together. Like tildens other
> turbots, if the connection worked poorly they would pull themselves apart,
> if it was benificial they would start doing their thing. Get the idea? Think
> about the things that could be put together. Comments?
>
> dennison
I beleive that highly modular, adaptive turbots are the most robust practical
BEAM design. I've also noticed that this approach replaces the Nv to biological
neuron linkage, and this is a good thing in my opinion. Rather than thinking
that
BEAM emulates insects, we should be aiming a lot lower on the evolutionary
ladder, at microscopic lifeforms.
These are natures turbots and beam-like life forms, and they also show the same
sorts of coupling we are discussing here.
Home