Alt-BEAM Archive
Message #04649
To: beam@sgiblab.sgi.com, Richard Piotter richfile@rconnect.com
From: Bruce Robinson Bruce_Robinson@bc.sympatico.ca
Date: Fri, 18 Jun 1999 20:43:15 -0700
Subject: [alt-beam] Re: MECI IV M (Was: Suitable for Framing)
Richard Piotter wrote:
>
> MECI IV M Functional diagram
>
> Microcore _______
> o-->o | |
> ^ | ------->|Nv 1,3 |
> | v ------->|Nv 2,4 |
> o<--o | | ________
> | MUX | | |
> Nv chain to | | | |
> lift the robot |OUT 1,3|----->|IN 1,3 |
> |OUT 2,4|----->|IN 2,4 |
> .-----+---------->|Nv 1,3 | | |
> | | | | | OUT 1,3|---->LEG MOTORS 1,3
> o->o->o->o--+---->|Nv 2,4 | | OUT 2,4|---->LEG MOTORS 2,4
> ^ |________| | | | |
> | ,->|CTRL | |REVERSER|
> | / |_______| | MUX |
> | / | |
> | / .------------>|REV 1,3 |
> | P-INIT | | | |
> '-----------+ +-->|--+----->|REV 2,4 |
> | | | |________|
> LEFT LEFT |
> OR FALL |
> RIGHT ONLY |
> |
> Fall triggers |
> |
> Reverse trigger
>
> Left fall trigger will trigger both reverses, while the reverse trigger
> only triggers one reverese
>
> Both left and right fall triggers will trigger the Nv chain as well as
> the first MUX.
>
> A right trigger must disable the walk reverse trigger???
>
> Obviously, I have work to do yet, but it's a start.
OK, between the diagram and your MECI IV page, I see (more or less) what
you're up to. I looks like you're after a forward gait, a backward gait,
and a "get up off the floor, dummy" sequence. Perhaps more. Those are
exactly the sequences I'll be using, except with a 3 motor testbed. So
my little idea may or may not be of any use.
This comes from human biology, specifically the "motor control" part of
our nervous system. Basically we control our muscles in a hierarchy of
nerve signals. At the bottom are neurons connected to individual muscle
fibres. At the top are individual signals that can set off a whole
sequence of highly co-ordinated motions. For an example, an experienced
driver has a single thought -- Stop!. No further conscious thought than
that, but an amazing sequence of motions take place to get our foot off
the throttle and onto the brake (and both hands on the wheel, and
possibly the other foot on the clutch). We have hundreds of such
sequences, built up by repetition and practice, all controlling the same
muscle fibers, all triggered with a single thought.
What happens between the single thought and the muscle fibers is a whole
set of complex interconnections that include several pattern generators
and feedback systems -- one for each complex motion we've learned well.
This is exactly analogous to having several different micorcore loops
(or possibly "polycore" loops), one for each motion we want our robot to
undertake. So instead of using MUX's to rearrange the signals to the
motors, or between Nv's, we have separate loops, only one of which is
switched on at one time.
The downside:
A whole bunch of extra Nv's, most of them idle at any one time.
Extra circuitboard space (or freeform volume) for the above.
Extra circuitry to activate one loop, and kill the other loops, and
combine the signals from the different loops.
The upside:
Easy to debug, and "train". You tune each loop to an optimum for its
particular function.
Easy to add new motions or patterns.
Such a scheme would not use a separate Nv loop for minor timing
adjustments (such as lifting the legs a little higher) -- just for
radically different motions, such as getting up off the floor.
I don't know if this will help at all (or even if it is practical --
that's this weekend's project). Hopefully it will stimulate a few
creative juices.
Have fun,
Bruce
------------------------------------------------------------------------
eGroups.com home: http://www.egroups.com/group/alt-beam
http://www.egroups.com
- Simplifying group communications
Home