RE: [Algorithms] Skeletal animation blending artefacts

[Tyson J. <twj...@sa...>]

Jumping spider?

Why not just move the whole spider up by the amount that its legs
penetrate?  Assuming everything is smooth (no discontinuities in
calculated interpenetration) your spider will look somewhat startled,
but won't look like it's a ghost.  Another thing to check is what is
happening on the root joint?  If it's moving away from the floor,
perhaps you could just interpolate the root faster so that the spider
gets its full height in time to avoid pushing into the floor.

I expect that aside from such quick and dirties, IK is the "real"
answer, isn't too hard (but do get a book :), and is fast *provided* you
have a good and well optimized math library.  IK tends to expose slow
functions in the math library once you get above 2 bones.

-----Original Message-----
From: gda...@li...
[mailto:gda...@li...] On Behalf Of Awen
Limbourg
Sent: Friday, December 06, 2002 8:33 AM
To: gda...@li...
Subject: RE: [Algorithms] Skeletal animation blending artefacts

Maybe you could compute a single 'spider-shoulder' rotation
transformation
to force its (bad, very bad) foot above the floor ?
(=>a single joint IK operation)

-----Message d'origine-----
De : gda...@li...
[mailto:gda...@li...]De la part de Jon
Jansen
Envoye : vendredi 6 decembre 2002 15:23
A : gda...@li...
Objet : [Algorithms] Skeletal animation blending artefacts


I'm using a simple time-driven interpolation system to give smooth
transitions from one skeletal animation to the next. However, it's
causing
rather unpleasant artefacts for certain transitions on certain models.

For instance, when our spider changes from idle to attack, the spider
can
be seen to momentarily (but smoothly!) disappear part-way into the
floor.
The problem is caused by the fact that I'm using simple linear
interpolation for the positional component of each bone's transform and
a
quaternion slerp for the orientation component.

It's easiest to illustrate if I do it in terms of simple trig. Suppose
the
height of the spider's body above the floor is 'h', and that the spider
has
a single leg of length 'l' which makes an angle 'theta' with the
horizontal. As animated, the spider's leg always makes contact with the
floor, so the artist has effectively arranged for:

	h = l * sin(theta)

Now suppose I'm blending between two animated configurations using a
blend
parameter b. If the configurations are:

	h1 = l * sin(theta1)
	h2 = l * sin(theta2)

Then the blending produces:

	h = b * h1 + (1-b) * h2

and

	theta = b * theta1 + (1-b) * theta2

But whoops, the constraint h = l * sin(theta) no longer holds true, and
the
foot penetrates the floor by a distance given by:

	b * h1 + (1-b) * h2 - l * sin(b * theta1 + (1-b) * theta2)

So (I think) I understand the cause of the problem, but the only
solutions
I can think of are either unpalatable or just plain unworkable:

1) Runtime IK (yikes!)
2) Ask our artists to produce transition anims (but the whole point of
blended transitions was to avoid this, and also to allow any animation
to
be interrupted by another at any time)
3) Ask our artists to modify the original animations to make them more
amenable to blended transitions (I'm not even sure I know what this
ought
to involve)

I'm beginning to think I can't have my cake and eat it...



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
GDAlgorithms-list mailing list
GDA...@li...
https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
Archives:
http://sourceforge.net/mailarchive/forum.php?forum_id=6188