Re: [Opensg-core] Animation Framework

[Marcus L. <ma...@ya...>]

Antonio Bleile wrote:
> Hi,
> 
> as we are collaborating with the local technical university
> and as we're offering some master thesis, we also proposed a
> thesis (with Dirk's authorization of course) that would resolve
> the animation issue in OpenSG, the title is "An Animation
> Framework for OpenSG". If there's time remaining, the solution
> also contains a plugin for 3ds and/or Blender that exports
> animations. The description of the thesis contains an
> explicit commitment to the Open Sourcness of the results and the
> collaboration within an  international and distributet team of
> developers (that would be you!).

Neato!

> We now found a young computer science student who is interested
> in this topic, and he's ready to go. But before taking off I've
> a couple of questions:
> 
> - Do you have some concrete ideas on how an animation
> framework should look like in OpenSG? Are there any
> papers/references other SceneGraphs where to learn from?

WildMagic/NetImmerse/Gamebryo is a set of similar SGs (created by the 
same guy)
that do animation.

Much of it is pretty good. The commercial variations have all the 
goodness you can expect (particle systems, skinning, keyframes, custom 
character animation cycle blending). Those things work as separated 
subsystems, which I think is good. They should be coupled very lightly 
to the geometry and to the clock.

It distributes time by traversing the graph. Every node has an 
update(float)
call that takes the time, with some flags that you can use to tag a node 
(and thus it's children) as time-independent. It is perhaps not optimal 
(a bit too tight coupling and too much graph traversal for my taste), 
but it makes it easy to add time-behaviour locally.

In our system, we have every "object" that need extra updating being 
registered to a "updater" when it's created. (By checking the type 
dynamically.) The order of update is simply the order of creation, and 
this has worked surprisingy well for almost three years now. The time 
(both absolute and relative) is fetched from a singleton (that provides 
different clocks for each aspect).

Also, OpenSceneGraph probably has something that works for them. Crystal 
Space is also more of a complete "engine" so there should be 
time/animation in there. A survey of scene graphs (and modelling 
software, if possible) might be worth it, unless we can decide on a 
method beforehand?

> - Where should he work on? In OpenSG 2.0 or OpenSG 1.8? I
> personally prefer 1.8 as I think we won't use 2.0 anytime
> soon.

I'd prefer focus on 2.0, so that it might be pushed far enough to ship, 
but I can live with 1.8 too. Best solution: make it lightly-coupled and 
thus adaptable to both.

> - What about Collada? Collada contains animation description and
> as far as I know, a Collada reader will be available in 2.0 ?
> 
> - I need some help from you guys taking care that the result
> is satisfying for all of us (a kind of tutor/mentor). Any volunteer?

I can give comments and offer my experience. But the design work has to 
be done by someone with enough insight, "feeling" & experience with the 
problem from the beginning. Either a bright thesis student, the OpenSG 
community and/or core-devs.

That's probably the trickiest part with any thesis work involving 
design. I think we ought to be lucky if we get _something_ workable if 
we set a "blue sky" target (I'm might be a bit pessimistic). OTOH, 
working after a spec, such as VRML (or something like Maya/VTK) might 
prove more fruitful (the focus would then be implementing some animation 
classes and have them work in a cluster).

Also, I recall Dirk having second thoughts about something as 
complicated as Maya's node-system, so a simpler solution should probably 
be the focus.

Cheers,
/Marcus