arms by trial and error

[Avery Andrews 931212.]
(John Gardner (931128.2000 EST))

Now that John Gardner is back and hopefully listening:

> It is fairly easy to set up a
>kinematic simulation of a given geometry and then 'construct' by trail
>and error, a way to map preceptions (am I using the right terms
>here? I don't want to get Rick all excited) to joint torques in such a
>way as to reduce errors.

Another (and, I suspect, more accurate) description of Bill's method
than `trial and error' would be `think about the concrete case at hand
and find ways of exploiting its special properties to produce a simple
solution'. In the case of the arm, for example, the elbow-flexion df is
the only way to to vary the shoulder-wrist distance, so there's no need
to mess around with Jacobian components, which sometimes point in
useless directions anyway.

PCT says that control will be easier to achieve if the controller has
the right perceptions - figuring out what these are is not necessarily
so easy.

Avery.Andrews@anu.edu.au

[From Rick Marken (931212.0900)]

Avery Andrews (931212.) --

PCT says that control will be easier to achieve if the controller has
the right perceptions - figuring out what these are is not necessarily
so easy.

What more can one say? Beautiful. Someday this will surely by chiseled
over the entrance to departments of robotics (which will be in the
school of "purposive systems").

By the way. Is La Trobe University anywhere near you? That's where
Donald MacPhee is located. They didn't put his e-mail address in the
article. The University is in Bundoora, Victoria, which might not
be part of cyberspace yet. I would like to get in touch with him
to get some ideas about how to do the "directed evolution" simulation.
Then I can stop ranting and start programmming and Aunt Mary might
like me again.

G'day

Rick