Alt-BEAM Archive
Message #05663
To: beam@sgiblab.sgi.com
From: Bob Shannon bshannon@tiac.net
Date: Thu, 12 Aug 1999 10:55:26 -0400
Subject: [alt-beam] Re: LEM videos now on-line.
Sean Rigter wrote:
> Congratulations Bob,
>
> Your whole LEM project (ie the physical design, shape and symmetry, sensors, the
> documentation and video) shows great finesse. Since so far there have been only
> kudos, I hope you are ready for some critical comments.
You mean some technical questions, and discussion of the design!?
Fantastic!
> I found the concept of a net linear displacement (ie the useful motion to move the
> LEM from point A to point B) of a triaxial wheeled platform difficult to grasp
> until I realized it is similar to Steven Bolt's 2 wheel design of a pivoting motion
> in which one wheel turns while the other is stalled. In Bolt's design, the 2
> unidirectional motors are alternately turned on (in one direction) for a 180 degree
> pivot around the stalled wheel resulting in a somewhat wobbly but on average linear
> motion while the platform is continuously spinning around it's center (in the same
> direction). This pivoting linear motion can be approximated with a conventional 2
> wheel platform which reciprocates (not spin) around it's center. The main advantage
> of Bolt's design is that it can reverse by turning on both motors and spinning in
> place around it's vertical center and can therefore "back up" from walls and
> corners. A conventional 2 wheeler with unidirectional motors can reverse only by
> keeping one motor on for 360 degrees instead of 180 degrees and which does not allow
> it to "back out" of a corner.
Not quite right.
LEM's 'Horse' layer is a very capable mobile platform even without the 'Rider'. Without
the Rider board to disable one of the three Bicore's in turn, it still moves quite well,
but
not in exactly the manner to seem to be describing here.
What happens is that the Bicores start to move, lets say all in the same direction. We
store this
energy in rotational inerita. At the point where one of the Bicores reverses direction
before the
others, this uniform rotational inertia gets transfered into linear motion. I think
this is a bit different than reciprical motion with 2 coaxial wheels.
Because no wheels share a common axis, this design will not roll down-slope even with
the motors unpowered. Without using gearmotors, coaxially mounted wheels will. Part of
what LEM does, is to use the transfer of rotational intertia to linear motion based an
asymetries in how that rotational torque is applied to the chassis. This design can
turn around in place, but it does not need to in order to escape from a corner like
Bolt's design, it will simply start moving in a different direction. I dont see any
clear advantage in either case.
(Actually, I beleive that the photopopper design uses this form of motion before Steven
Bolt, its really only a matter of how much capactiance you charge to what voltage that
sets the swing angle. My most 'effieicnt' (fastest) photopopper design turns 1bout 120
degrees per step, pivioting around the unpowered wheel.)
As for the slipping and sticking of the wheels, please notice the sounds LEM produces.
This is
caused by many small splines on the motor shaft gears being used as wheels. Because
these splines are parallel to the motor shafts, the 'wheels' can be easily dragged along
the axis of the
motor shaft with very little friction. LEM 'skipps' over the surface its moving on.
As the other 2 motors also have the same type of wheel, there is a fair ammount of
vertical, high frequency viberation of the chassis. Should a unpowered wheel catch on
something, its quickly nudged over the obstruction. Also the design rarely moves in any
given direction, so as a result
of all these factors, frictional losses are quite managable overall.
Using gearmotors and larger wheels, it would be critical to adjust the timing of the
motor drivers to match the mechanical resonance of the chassis in order to acheive the
same form of motion that LEM is using. Part of the whole idea of LEM was to develop a
way to apply non-gear reduced motors to drive a 'bot over terrain normally thought to be
impassable to direct drive designs.
The key to this, is to store energy, either in some form of suspension, or in this case,
as rotational inertia, and then apply it in small bursts of motion with more energy than
the motors could apply directly. It clearly works, but I admit it could be made much
more efficient by adding a more conventional suspension system to each motor.
> So in a 3 wheel platform one can approximate Bolt's linear motion with two wheels
> spinning in the same direction while the third is reversed causing pivoting around
> an off-center pivot point and resulting in a net linear displacement.
That seems to be a very loose approximation, but basically correct, but only for a tiny
fraction of the time that LEM is moving. The motors reverse direction much faster,
keeping the LEM 'aloft'
in a fast 'skipping-like' motion.
> While it was possible to conceive of such linear motion, I had problems
> understanding the frictional losses in such a 3 wheel design so I decided to build
> one. I used 3 gear motors mounted horizonally with 1 inch diameter toy wheels
> attached on the perimeter of a 5 inch circular platform. I used a simple tricore
> with the 3 outputs connected to 3 bridge circuits. This results in a simple sequence
> of 2 motors Forward (F=1,2)and one motor Reversed (3), followed by F=2,3 R=1 then
> followed by F=3,1 and R=2 . The Tricore timing was adjusted to a rotation of
> approximately 120 degrees around the pivot point
This does not approximate LEM at all. LEM's basic form of motion does not require that
I control the timing of the Bicores in any way. You may be confusing functions of my
Horse
and Rider layers here.
I only disable a single Bicore at a time in order to save power, but this does not
effect the basic
gait of the machine. The translation of rotational inertial to linear motion is only a
function of the relative timing between the bicores themselves. No Rider board is
needed at all, so the sequencing of the motors is not as you described above. Imagine
that each motor is driven by a free wheeling Bicore using the standard .22 uf caps,
etc. They oscillate rather quickly.
The speed of the Bicores is such that they may reverse several times per linear 'step',
so the thing actually sort of skipps and glides over smooth surfaces, but digs in and
crawls over rougher surfaces or small obstacles. On a really smooth surface, you can
see this by giving the thing a good push with your hand, it spins and glides like an
air-hocky puck, regains traction, and comes right back at you, acting really mad about
it too, in a very life-like way. I cannot see how you can
be close to this with gearmotors.
> I found that net linear motion does results in my 3 motor design but is very
> inefficient with a lot of wheel dragging and slipping resulting in high frictional
> losses compared to, for example, Bolt's 2 motor design! These losses are caused by
> the "toe in" of the 3 wheels which oppose all net linear motion. Even if one wheel
> is held perfectly stationary causing the platform to rotate about that relatively
> low friction pivot point , the sideways dragging and slipping between wheel and
> surface (ie friction losses) of the other 2 wheels is very high.
>
> So my question are:
>
> Am I missing something?
Yes.
First, I have to admit that LEM is not as efficient as a 2 motor rover. Nothing is,
except a unicycle in theory. No walker, nor any other design will ever match a wheeled
rover for efficiency, thats why we invented bicycles after all.
But LEM moves a lot of weight over a wide range of terrain without gear reduction. It
does draw some current, but it will run a long while on those 280 maH NiCads, and its
tossing solar designs aside the whole time. So I would not call LEM particularlly
inefficient either.
But what you are missing is simply that the wheels are not intended to have solid
traction while its in motion. I've experimented with a number of wheel designs on LEM,
and even high friction rubber wheels work better than I had expected. The small gear
splines however, work far far better than I could have imagined. The secret here is in
the sound it makes.
LEM is not a true roller at all, thats why it does not roll down hill when it sleeps.
Try this with
a direct drive, coaxial motor design.
I think that the really poor frame rate of the videos is causing some confusion here.
The NTSC video tape alreay looked slightly choppy because LEM really moves much
faster, and in a much more complex way than the compressed videos show. After
compression,
even at the 128K/second rate, it looks (to me) like very bad clay-mation.
The original, uncompressed AVI file is about 142 megabytes, so you can see that a lot of
detail is
lost by the time you see the RealVideo versions over the web site.
If anyone really wants to see the full 142 meg AVI version, I can try to make it
available.
> Do you observe similar "wheel/surface" drag and slip by peeking on the "skirt" of
> the LEM? (it is impossible for us to see this on the video)
Yes, I tried, but it moves so quickly that this was impossible to capture on NTSC video,
much
less any compressed form. I suppose I can make you a very short AVI file, with a
close-up of
one wheel-footie-thing, and see just how bad it turns out. Maybe it will suggest the
details of
the wheel to floor interaction, but I really think that the sounds it makes tells you
whats happening
here better than even the naked eye can do.
> In your opinion, would you say that the LEM design has an efficient linear motion (
> getting from point A to point B)?
Given its weight, and the work (expoloration) it does along the way, its nominally
efficient I think.
But its design includes a lot more than moving from point A to B, if that were the goal,
I would use
a different design, a conventional wheeled rover. LEM was designed for the form of
motion contest, where praticality was the stated goal. Turn LEM loose in your RJP, and
see how practical it is compared with convetional BEAM designs.
And its efficient enough to carry along a solar recharger system, giving it a greater
survival
space than a lot of other (more efficient) designs.
> What do you think is the efficiency of useful linear motion of your 3 motor
> triaxial LEM compared to a 2 motor axial + one caster platform controlled, for
> example, by a Unicore circuit which has a similar Horse and Rider (embedded
> Bicores) architecture.?
Ahhhh, now you have gotten to the real point behind this design!!!!
Its all about the control system design philosophy. If you design a geometric resonance
between the control system, sensors and effectors, your going to get something very
interesting.
My philosophy with LEM is that BEAM tech is much better applied if we try to emulate
much lower forms of life that is common. We compare BEAM designs with ants, clearly
absurd! We should be trying to build a starfish or similar design first. In these
creatures, we see no neurons at all, yet we have very complex, directed behaviors
emerging from an array of rather simple subsystems in a geometric design.
If efficiency is a goal, you could try a true roller design, if you constrain the
environment to smoother surfaces, or use larger diameter wheels and gear reduction. The
controller time constants would have to be altered, and I might reccomend spherical
wheels as well...
If robustness is more important (as in a predatory design) then don't be afraid to burn
the milliwatts as long as its getting work done fast. LEM does this well enough I
think.
> Do you think that by extrapolating the LEM design to a 2 motor or 4 motor design,
> would this result in a more or less efficient linear motion?
Its much more complex than this. What is your working definition of more or less
efficient?
If its milliwatts per centimeter over time, the number of motors is not the main
factor. If I used
a turbot-LEM design, I may have many motors, but few ever get power. Those that do may
have a good enough match between their gear ratios and mechanical advantges of the arms
that while very slow, its also very very efficient. In this case, it may indeed be more
efficient than a design with fewer motors and high frictional losses.
Personally, I think the most practical design might be a 4 motor tetrahedral turbot,
able to operate efficiently in any orientation, over most any terrain.
> I look forward to comparing my tentative results with your experience (but of course
> you have the right to remain silent - grin)
>
> regards
>
> wilf
Oh no I dont!
This is an open contest, to enter, I must disclose my design fully I beleive. Anything
less would
not be fair to the community.
Remember, its not a conventional roller design, thats why the motors are not coaxial.
Maybe its no model of efficiency, but its cheap and practical, needing no hard to get
(win)
gearmotors, yet it performs in a very interesting way.
Try to test the actual Horse design, with direct drive motors and totally seperate
Bicores.
You can have a lot of fun just testing out what all the different mappings between the
photodiodes and bicores do. It gets very complex, even without any 'Rider' at all.
What your testing right now does not seem to desribe LEM's form of motion very well.
Home