Re: Open source robot controllers

> We really need help for :

Just a comment: for most of these points, there's already something, but
it could be improved

> - Documentation! (A lot of work to be done there!) Overflow is very
> powerful, but its lack of documentation makes it hard to get started.

There's a lot to be done in documentation, but there's also a lot
already done, so you might want to have a look and see what's in there.

> - The Overflow core could be used on embeded devices. Instead of having
> a multitude of toolkits (shared libraries), we are working on a C++ code
> generator that generates the whole program & makefile. So, applications
> created with Overflow can be packaged into a single stand-alone program
> (without the need for XML, GNOME, etc.).

Actualy, the code C++ generator already works and the company I work for
will be using it for a part of their product in the next version (BTW,
no license problem, it's LGPL). 

> - Right now, Overflow networks can be "serialized" an sent via TCP/IP
> for remote execution. Maybe it could be useful for things like
> seti@home... We could have a nice client/server demo where clients
> download jobs from the server and returns the processed data...

> - Jean-Marc (the other project manager) works mostly on audio procesing.
> We had the idea to integrate Overflow to gdm / kdm (login manager) to
> add voice authentification. A lot of work still need to be done, but I
> think it could be a really nice application.

That's me;-) Yes, in case some are interested, the audio/DSP toolbox is
getting quite complete, as I wrote my whole master project (speech
enhancement algorithm) using it.


Also, I'm not sure I understand what you meant about Overflow being or
not like Simulink, but at least I can explain a bit how Overflow does
it. First (unlike, I think Simulink), an Overflow node can return any
type of Object (scalar, vector, matrix, neural network, IO Stream, ...).
From the GUI, you can plug any node into any other node, except that if
the output of one node is not what the other node expects as input,
there's a run-time exception. 

The idea between an Overflow node is that you "ask" for its output and
in order to do the computation, it (recursivly) asks its input nodes for
their return value. In that way, the final result is thus computed
recursivly by asking the last node for its output. As you see, the
original assumption is that nodes only calculate and have no "side
effect", but the same idea also works for nodes with side effects (like
controlling a robot). You can find a more details on that here:
http://freespeech.sourceforge.net/FreeSpeech/html/Overflow/overview.html

Now, about the idea that orocos would provide a framework and not allow
any other way to connect nodes, I think that would be simply implemented
as a layer above Overflow that would add restrictions on the
connections.

Last, the main Overflow documentation page is at:
http://freespeech.sourceforge.net/FreeSpeech/html/Overflow/doc.html

	Jean-Marc