[Avery.Andrews 930111.1905]
Reading around in the Osherson, ed. `foundations of cognitive science'
(1989) I came across the claim that feedback is to slow to solve inverse
kinematic & dynamic problems for fast movements. Where can I read about
why this claim is false or irrelevant (e.g., true only for certain
kinds of highly skilled movements that people practice enough to make
it plausible that they have elaborate feedforward schemes for).
Avery.Andrews@anu.edu.au
[Chris Malcolm]
Avery Andrews writes:
I came across the claim that feedback is to slow to solve inverse
kinematic & dynamic problems for fast movements. Where can I read about
why this claim is false or irrelevant?
It's true for assembly (arm-hand type) robots. See any intro robotics
textbook. One of the reasons why assembly robots can't compete with
human assembly speeds is that solving the inverse kinematics at the
speed of fast human arm movements is just on the brink of being too
expensive computationally. Inverse dynamics is orders of magnitude
worse. But of course we are talking here about algorithmic solutions
based on mathematical models, and that is not the only way of solving
these things (e.g. Churchland's crab). In the end you come up against
imagination failure, i.e., "we can't think of a way of doing this, so we
think it is probably impossible."
[From Bill Powers (930115.1330)]
Chris Malcolm (930115) --
One of the reasons why assembly robots can't compete with human
assembly speeds is that solving the inverse kinematics at the
speed of fast human arm movements is just on the brink of being
too expensive computationally. Inverse dynamics is orders of
magnitude worse.
Are you saying that the designers of these robots are computing
inverse kinematics? They're not using direct feedback control of
joint angles?
Have you actually seen Little Man Version 2? It controls an arm
with 3 degrees of freedom without computing inverse anythings.
The only kinematic/dynamic computations are the ones that model
the arm's response to torques at the shoulder and the elbow. In a
real robot the arm itself would take care of that part of the
"model."
I must be misunderstanding what you're talking about.
Best,
Bill P.
[Avery Andrews 930116.1130]
(Chris Malcolm)
I came across the claim that feedback is to slow to solve inverse
kinematic & dynamic problems for fast movements. Where can I read about
why this claim is false or irrelevant?It's true for assembly (arm-hand type) robots. See any intro robotics
textbook. One of the reasons why assembly robots can't compete with
I actually did (by some people from the Uni of Wollongong, dated 1991, but
without quick illumination. Presumably the implication of what your
saying is that the armdemo, even running on a very fast computer,
couldn't keep up with a fast & practiced assembly-line worker, at least
if its degrees of freedom were more like those of a real arm). But
of course these people are presumably not developing the inverse
kinem-/dyn-amics from scratch every time the move: I would presume
that they find & develop a solution through constant practice.
Avery.Andrews@anu.edu.au
[From Chris Malcolm]
Bill Powers writes:
Chris Malcolm (930115) --
One of the reasons why assembly robots can't compete with human
assembly speeds is that solving the inverse kinematics at the
speed of fast human arm movements is just on the brink of being
too expensive computationally. Inverse dynamics is orders of
magnitude worse.
Are you saying that the designers of these robots are computing
inverse kinematics? They're not using direct feedback control of
joint angles?
They are doing both. Each motor is PID speed and position controlled by
direct feedback of joint angle (actually motor rotn). If well done this
by itself is sufficient to get the arm to move on a smooth and
reversible trajectory from A to B, with all motors starting and
finishing together, along a curve close to the minimum energy curve,
known as joint-interpolated motion.
But if you want the end of the robot to move in a straight line you need
to solve the inverse kinematics. Because it is hard/expensive to do this
within the usual 10/20 msec motor control loop cycle times, the usual
procedure is to approximate the straight line out of a lots of little
joint-interpolated curves, by calculating enough intermediate points
along the straight line, and feeding the next set of target joint angles
into the motor control loops before the motors have got to the previous
target. It becomes hard/expensive to do this fast enough when robot arm
speeds approach fast human arm speeds. "Expensive" means the reluctance
of the marketplace to spend much more on a robot control computer than
on the electromechanics of the robot. This "moving carrot" kind of
hierarchical control should be familiar to PCTers!
Have you actually seen Little Man Version 2? It controls an arm
with 3 degrees of freedom without computing inverse anythings.
3 degrees is trivially easy to do almost any way you like. It's the last
two degrees of freedom (5 & 6) that create the computational
difficulties. I know you can solve these movement problems without
inverse kinematics. I was responding to a poster wondering where the
"feedback too slow" came from. There is such a general piece of folk
wisdom in the domain of industrial assembly robot arm design, and I was
explaining where that idea came from in that domain, where it naturally
reinforces the piano-player argument. It is only relatively recently,
say in the last seven years, as the limitations and expense of the
"obvious" Cartesian/Newtonian approach to arm control became obvious to
roboticists that they started looking for alternative solutions. The
general bias of engineers (who for some reason seem to be even "harder"
minded than physicists) against the "soft" sciences has meant that only
a few went so far as to look carefully at biology, let alone AI and the
systems theory/cybernetic roots of AI. In biological and control
literature "Behaviour: the Control of Perception" shows such an unusual
grasp of the architectural possibilities of hierarchical control that it
it has been a minority interest among some of the
biologically-interested roboticists ever since its publication.
Chris Malcolm
[From Rick Marken (930117.1130)
Chris Malcolm --
But if you want the end of the robot to move in a straight line you need
to solve the inverse kinematics.
On what basis do you make this claim? Assuming that this is true for your
robots (for some reason), do you imagine that this is a law of nature;
that this kind of straight line movement requires the solution of
inverse kinematics? Have you ever actually TRIED building a robot
based on PCT? If so, and you still come to the conclusion above,
then I think it would be VERY important for PCT theorists to know
precisely WHY this is so.
Powers said:
Have you actually seen Little Man Version 2? It controls an arm
with 3 degrees of freedom without computing inverse anythings.
You reply:
3 degrees is trivially easy to do almost any way you like. It's the last
two degrees of freedom (5 & 6) that create the computational
difficulties.
So can we expect the PCT revolution to begin when the Little Man
also controls the roll (4) pitch (5) and yaw (6) of a baton held
in his hand? I would guess that after all the work to program these
changes (the main work being simulation of environment; the
torques of the baton and whatnot) the AI, robotics, and motor control
community would look at it and say "yeah, but your doing it with a
little man; most baton twirlers are girls". This critique is as cogent
(and misses the point as grossly) as does the observation that
only 3 df are controlled in the current demo.
Best
Rick
[Avery Andrews 930118.1024]
Chris Malcolm 930117
> It becomes hard/expensive to do this fast enough when robot arm
> speeds approach fast human arm speeds. "Expensive" means the reluctance
For practiced or unpracticed movements, and with fixed or variable
loads? These seem to me to be crucial questions about arm, which seems
to be the leading-edge technology in this area.
Avery.Andrews@anu.edu.au
[Avery Andrews 930119.1922]
Maybe another contributing factor to the `feedbback too slow' story is
the work by Rack and others (who seem to know what they're donig) arguing
that the spinal reflexes are too slow and have too little gain to
be much use in compensating for short-term disturbances.
Hopefully this doesn't imply that they're not useful for dynamics - it
is perhaps significant that amphibians, who lack the dual alpha-gamma
efferent system, are not noted for their acrobatic abilities.
Avery.Andrews@anu.edu.au
[From Chris Malcolm]
Rick Marken writes:
Chris Malcolm --
Powers said:
Have you actually seen Little Man Version 2? It controls an arm
with 3 degrees of freedom without computing inverse anythings.
You reply:
3 degrees is trivially easy to do almost any way you like. It's the last
two degrees of freedom (5 & 6) that create the computational
difficulties.
So can we expect the PCT revolution to begin when the Little Man
also controls the roll (4) pitch (5) and yaw (6) of a baton held
in his hand? I would guess that after all the work to program these
changes (the main work being simulation of environment; the
torques of the baton and whatnot) the AI, robotics, and motor control
community would look at it and say "yeah, but your doing it with a
little man; most baton twirlers are girls". This critique is as cogent
(and misses the point as grossly) as does the observation that
only 3 df are controlled in the current demo.
I agree. But the kind of engineering types who build industrial robot
controllers are not going to even bother examining the detail of any
demonstration of a supposedly better way of controlling robots which
only handles 3 DoF, since they know that 3 DoF can be done very
computationally cheaply by a great variety of methods which fail
horribly with 6 DoF, and students keep coming up to them with very
wonderful methods which work a treat in 3 DoF and bog down horribly in
6. "Very interesting, come back and show me it handling 6 DoF, and then
I'll be interested enough to bother trying to understand it" is the
standard response.
Chris Malcolm
[Avery. Andrews 930121.700]
I don't think that sending out a list of embarassing quotes to 50 big
names is the right way to go. After all, it's not the big names who
are the worthwhile targets, but their prospective students, and assorted
dissatisfied mavericks. Priority one is to get a useful document
available online for CSGNet, so that people who are intrigued by what
we're up to but not sure that it isn't bull can get some easily
verified hard evidence that many establishment figures are seriously
confused. Then perhaps some kind of article somewhere.
Re 6 DoF, has anyone analyzed the DoF of a set of octopus tentacles,
I wonder? (and they can learn to open jars with them ...)
Avery.Andrews@anu.edu.au
[From Bill Powers (971217.0504 MST)]
In _Nature_, 4 Dec. 1997, there is a "letter to Nature":
Bussetini, C. Masson, G.S. & Miles, F.A.: Radial optic flow induces
vergence eye movement with ultra-short latencies. Nature, _390_, 512-515
(1997).
The authors had people watch screen presentations with varying amounts of
radial flow, producing the illusion of approach or recession. The eyes
showed appropriate vergence movements (converging or diverging). The
reaction time was consistently measured at about 80 milliseconds. Each eye
moved opposite to the apparent direction of flow primarily in the nasal
hemifield.
This was apparently an open-loop measurement that depended on a motion
illusion, but it shows that visual-motor latencies can be as short as 80
milliseconds.
Best,
Bill P.