[From John Gardner (950114.10:30 EST)]
Hi all, long time no post. I don't always read everything that shows up on
this group but when you start dragging the roboticists and engineers through
the virtual mud, I simply can't resist 
I started contributing to this list when detailed discussions about the hwys
and wheres of robot control systems became and issue and I seeme to recall
some interesting and fruitful discussions. I think that, once again, I might
be able to contribute something here.
I think that there are some misconceptions about what control engineers 'do'
and why robot control schemes have been developed the way they have. Let me
say at the outset that I am attempting to present these point in a
non-argumentative way. In other words, I'm not saying that we're "right" or
"wrong" in doing these things-- I'm just trying to elicidate what it is we do
and how we (controls/robotics engineers) do it.
First off, I think it's wrong to assume that all research in robotics is
attempting to emulate human or other biological behavior. From an
industrial/engineering standpoint, we're trying to move stuff around, perform
operations on parts (drilling, grinding, etc) or perhaps trying to move the
robot itself. To say that we're trying to 'copy' human behaviour is a bit
like saying that an automobile is an attempt at copying (and then,
understanding) a horse. Of course, with that interpretation, the car is an
utter failure but.....
Anyway, there *are* a lot of robotics people who claim that their research
will provide valuable clues to understanding biological-based motion, and
perhaps they are the target of your attention,and perhaps correctly so. I
just wanted to point out that there are a lot more (IMHO) robot/control
engineers who are just trying to make machines.
As a graduate student at Ohio State, I worked on a project in which we
designed, built and controlled a six-legged vehicle (not techincally a robot,
since it was 'driven' by a human). At a conference I took some heat because
we controlled the machine in ways different from how insects walk. At the
time, I didn't know how to responde, so I didn't but in retrospect, I believe
that the situation is similar. Of course insects walk differntly-- the scale
is so completely different (our machine was about 3000 lbs and over 3 m long)
The fact is, it walked and did better than any other legged vehicle of that
scale before (or since, as far as I know).
There's considerable energy expended in this forum about the 'inverse
kinematics' and other mathematical tools of the 'traditional' robot control
scheme. I've talked about this before, and I agree that it's highly unlikely
that any portion of a biological system's resources are devoted to the
solution of trignometric equations, finding values of angles and desired joint
rates. I don't know any engineer who *does* believe that. However, I would
argue that these approaches represent perfectly valid and appropriate
approaches to the problem of controlling *machines*. Let me explain.
When you take chunks of steel and aluminum and connect them together with
pivoting joints and mount motors at those joints (sliding joints too), you get
unbelievably complex, nonlinear dynamics. In fact, much of the resulting
system behavior becomes extremely non-intuitive. Simple assumptions like:
"when I push in this direction, I will get motion in this direction" don't
apply. Equally important, characteristics of the system which you learn for
one 'neighborhood' of it's 'behavioral space' (by this I mean, a certain
'pose' and velocity of the mechanism) may (and probably will) be very
different from characteritics in another neighborhood. This is why we
(engineers) prefer the mathematics of Jacobian matricies to relate desired
hand motion directions to command motor rates.
Bill has mentioned in the past that the simulations used in your
demonstrations are kinematic in nature: mass and rotary inertia are
neglected. While I agree that the results to date, as I understand them, are
impressive (why do you think I still lurk around here?), the mass effects will
make things *very* interesting.
A related point which has occurred to me is that part of this discussion may
be getting at some fundamental differences in design. In other words, the
traditional robot is designed with n Degrees of Freedom (DOF) and n motors,
usually colocated with the joints and feedback sensors. It doesn't take a
Ph.D. in human anatomy to see how very different that is from biological
systems. I wonder if evolution hasn't arranged some of the muscle groups in
such a way to make control (perhaps *perceptual*) control easier. It
certainly seems easier than the approach we use in hardware right now.
So, let me leave you with this suggestion. I've seen the 7 DOF arm simulation
that Bill has written but I just can't see myself spending the time to thrash
through the dynamic equations of motion for such a beast (I hate 3-dimensional
coordinate tranformations, they make my brain hurt) BUT-- I have a two-link,
two-motor direct-drive robot arm, controlled by a 486PC, programmed in "C"
which is always looking for interesting experiments. The robot works in a
horizontal plan, positioning it's end-effector (a plotter pen) in a circular
workpace that's about .8 m in diameter (I use only one quadrant of the
workspace. The first link is 0.39 m long, the second is 0.42 m. I'm using
NSK megatorque motors which are capable of some pretty hefty torques (motor 1-
40 N-m, motor 2 10 N-m). It's pretty massive and can move pretty darn fast
when it 'wants' to. What makes it interesting is that the direct-drive nature
of the robot makes the multi-body dynamic effects very significant. Decent
control of this beast if very difficult.
The bottom line is that I'd be interested in implementing a proposed control
scheme and report the results. I can't promise what kind of time frame we'd
be dealing with, but I think that I might get lucky and find a student willing
to 'noodle around' with the robot for me.
So, enough from me for now, I have to get back to work and go back into the
woodwork, where I belong. I look forward to reading the follow-up posts.
John Gardner