RE: Modelica Conference 2002

On Wed, 6 Feb 2002, Tiller, Michael (M.M.) wrote:

[...about Modelica...]
> > >   My guess is that you are really interested in a simulator, correct?  
> > 
> > A simulator indeed, but more than that: code that can be used on-line
> > in a real system.
> 
> Do you mean for doing embedded system code generation?  Are you
> talking about logical control or continuous behavior.  

I'm talking about the dynamics algorithms of robots, to be used by the
on-line controllers. Preferably with the ability to switch control
laws on-line, which is needed, for example, for controlling a walking
robot (during subsequent phases of a step).

[...]
> Well, the full Modelica language would require the ability to solve
> continuous equations and respond to events based on conditional
> expressions (involving continuous and discrete variables).  That is
> the basic part.  The complete semantics include some complexities
> for doing things like initialization of discrete controllers so that
> the controller starts in "steady state".  Also, for systems of DAEs
> there is the issue of index reduction (this need not be handled
> directly if you use a DAE solver, but it is my understanding that
> the general numerical solvers are really only able to go up to
> index-2 DAEs while the Modelica language allows arbitrary index
> DAEs).  There are several other issues but these are the ones I
> remember off the top of my head.

Ok, I see and know these problems. But for use in a controller, we
don't need these solvers: nature does that job for us :-) I just want
to reuse as much as possible from the modelling and specification
efforts.

[...]
> > That I understand :-) But I still sort of dislike its `feature' that
> > extra things (such as 3D object information) have to be given in
> > ``comments''.
> 
> Well, that isn't quite a fair statement.  

Sorry about that :-)

> The 3D object information
> is not stored in a comment.  In Modelica (as in most languages), a
> comment is a free-form string that only provides documentation.  In
> Modelica both 3D representations and 2D representations are stored
> as *annotations* which have a specific syntax, allow for structured
> data representation and can be associated with specific instances or
> definitions.  

Okay, but just calling things by another name doesn't bring us any
further :-)

> > And how good would Modelica be as a general controller description language?
> 
> It isn't clear to me if you are talking about continuous controllers
> or discrete controllers.  

Hybrid controllers! That is, with continous and discrete parts. In
robotics, you quickly end up in this hybrid controller area.


> I would need an example of the kind of
> controller you are talking about.  

Walking robot: in each phase of the step, one uses a continuous
controller, but there are many (more or less continuous) transitions
between phases: swing to heel contact, toe contact to release, etc.


> The purpose of Modelica is to try
> and capture a reasonable set of behavior types in a single language.
> If it can't be used as a general controller description language, I
> would be very interested (as would the rest of the design group) in
> understanding why and seeing if the issue could easily be addressed.

That would be a nice project :-) I am on the outlook to participate in
it.

> We are not interested in creating some all encompassing, monolithic
> language but I see no reason why the application you describe would
> not fit within the existing framework.

That's the reason why I like XML: it's never a monolithic thing, not
because of its inherent properties, but because there are so many
tools available to transform from and to XML.

> Are you familiar with Simulink and Stateflow?  If so, could you
> perhaps describe your application in the context of those tools
> (since I am somewhat familiar with them)?

I know these, but the reason I like Modelica is that Simulink leads to
spagetti code quickly, while Modelica is inherently much more
component-based and object-oriented. I mean, I see students connect
anything to anything in Simulink, while they should be able to do
that. A good interface (human, parser) to Modelica should make these
errors impossible.

Herman