Re: ORBit for robotics??

[Mårten Björkman <celle nada kth se>]

On 23 Oct 2002, Michael Meeks wrote:
> 	Ok; so ORBit2 should give you far faster compile times - especially if
> you use pure C. There are embryonic - but mostly working C++ bindings in
> the 'orbitcpp' CVS module. ORBit2 also has some nice python bindings
> coming along.

C++ is nice, of course, but stability of really important, so we
might prefer a stable C binding from a less stable C++ one. On the
other hand, some C++ freaks in this lab might not agree. Personally,
I prefer assembly, but few people seem to agree.

> 	If you have problems with exotic non-poll driven I/O then I'd recommend
> using a helper thread to do that badness, and re-vector the input stream
> via a socketpair to the main thread - we do that in GnomeVFS to good
> effect for example.

I believe, that is similar to what we have been doing so far, at
least for many of the input devices. For larger memory structures
(like complete images) we use shared memory though. That's one of the
reasons I'm a bit worried about threads.

> 	As for video transport; I've no idea how fast you expect that can be;
> but I doubt we'll be markedly faster than ACE - if you have lots of
> data.

I agree. We are well aware of such limitations and subsample the images
whenever possible, before sharing the data. That's why people in this
lab refer to Vision as Wishion. Everyone likes to use Computer Vision
for Robotics, but there are just so many obstacles.

Thank you very much,

Mårten Björkman