RE: Modelica

On Mon, 30 Jul 2001, Tiller, Michael (M.M.) wrote:

[...]
> > But for other modelling parts, Modelica is not suitable (I think): I
> > mentioned already trajectory specification and component configuration
> > in another email.
>
> By trajectory specification, do you mean specifying inputs that are
> functions of time?

I mean describing how your robot should move. Trajectory generation in
itself is a very broad and complex problem, if you involve smoothing
and blending of pieces of motion, or distributed motion generation and
control, etc. I don't find any dedicated features for this in
Modelica.

> It not clear to me what you mean by component configuration.  Do you
> mean describing assemblies of machines in terms of their components?

For example (this is the `mechatronic modelling' I spoke about
before), but certainly not only that. I indeed had something larger in
mind: in my view, a modern robot control system consists of several
computer systems controlling different parts of the robot system (the
`classical' robot arm or mobile platform; `intelligent' sensors such
as vision systems and autonomous localization modules; a
planning/trajectory generation componente etc.). The specifications
for each of these I am currently developing on the web page (far from
finished!).  Hence, in general, there will be several more or less
independent software tasks running (``threads,'' ``agents'',
``components'', ``holons,'' or whatever name they are given), and each
of these tasks has a certain `state'; this state has to be
initialised, (re)configured, sent to other components for diagnosis or
information exchange, etc. All this complex data management (as you
aptly called it) requires a `language', and XML seems to be the
appropriate settings to make this language.

[...]
> I would point out that the Modelica Association is open.  This is to
> get a wide range of opinions on the functionality of the language
> and to make sure that updates are agreed upon and known in advance
> to those affected by them.

I have had in mind for a long time already to participate to the open
discussion and proposal process and attend some of the quite frequent
workshops. But I don't yet have a sufficiently precise and detailed
proposal ... I hope this will be one of the early results of Orocos :-)

[...]

> > > First, it is clear from the data what the types of the elements are
> > > (i.e. not need to look back at the schema to find out whether
> > > something is a real or a string).
> > My evaluation: this is a good feature for human interaction,
> > but irrelevant for the software.
>
> An important difference, perhaps, between XML and Modelica.
> Modelica is, first and foremost, a human readable language.

I think we reached an important conclusion here, which, as usual, is
so obvious once you read it :-) (But I must confess it wasn't all that
clear to me until now.) So the question seems to be turned into `How'
to best integrate Modelica, instead of the earlier question `Modelica
(exclusive) or a new XML language'.

[...]
> I should point out that for reasonable small amounts of data, this
> array syntax is very nice.  For larger sets, readability is not so
> important and instead binary files can be used (MATLAB format).

This is another `classical' discussion: is Matlab to be trusted enough
not to change the binary Matlab format too often, for the
sole reason of forcing customers to update? (They learned a lot from
Microsoft in this respect...)

[...]
> The "templates" functionality can be used both in a compositional
> context and an inheritance context.

Interesting remark! (But I don't fully grasp the idea at first reading :-)

Herman