[Robotics] Itasc branch review

[Benoit Bolsee]

Hi Florian,
 
Thanks for this review. I'm surprised by the comment on one core being
completely busy during animation playback/record. I had exactly opposite
experience when testing complex armature: with old solver: 100% busy,
with itasc: 8% busy.
 
I can answer your question about pose flipping when you change a target
in 3D view: any modification of an object position causes the dependency
chain to be 'flushed'. This means that the armature, which depends on
the target, will receive a 'cache flush' request and the entire history
of the solver is cleared. As a result, the solver doesn't have a past
position anymore to compute the current pose, and starts from the rest
pose, causing possible flip. Once you replay the animation in full, the
cache will rebuild automatically and you get the correct dynamic again. 
 
I didn't find a simple solution to this problem yet. One possibility is
to use the current pose instead of the rest pose as the start point in
case past history is missing in the cache, but that would create a
dependency on future when playing an animation in loop. Another
possibility is to rebuild the cache in full each time the cache is
flushed, but that's not viable performance-wise. A third possibility is
to use the pose from Fcurves as a start point instead of the rest pose.
These Fcurves would be automatically updated (inserting key frame for
all bones) when modifying targets. This is probably the best option but
requires actions on Fcurves that I'm not familiar with.  
 
I'll do some experiment as soon as possible. For the time being, you
need to play the animation each time you change something on the
armature.
 
/benoit 
 

-----Original Message-----
From: robotics-bounces at blender.org [mailto:robotics-bounces at blender.org]
On Behalf Of Florian Lier
Sent: samedi 26 septembre 2009 10:42
To: Blender and Robotics
Subject: [Robotics] Itasc branch review


Hey all,

finally I found some time to compile and test the itasc branch.

- Some notes - 

Test system:

Ubuntu 9.04 32 Bit
Intel QuadCore 3,2 Ghz
4 GB Ram
NVIDIA 8800 GTX 512 MB with NVIDIA prop. driver

 ":("
1) With Compiz/3D effects enabled the game engine seems to run pretty
unstable
2) Compiz also seems to mess up the menues like "Armature" "Bone" or
"Contraints" Menue (screenshot: www.icram.de/itasc_scale.png)
3) When recording an animation and replaying it one core is completly
busy (100%)


":)"
1) The menu-settings for the IK are pretty nice and clearly arranged
2) The solver works very well and seems to be smarter/smoother than the
"normal" solver
3) The scons installer works flawlessly


That's the my first impression on the current itasc branch. I compiled
it myselft - I didn't
use the binaries provided by benoit.

Here's a small video showing the demo: http://vimeo.com/6761002


A general question: When moving the IK chain controlled by an EMPTY
(target) in 3D View 
the solver sometimes "jumps" from pose A to pose B. During replaying and
animation the solver
doesn't jump that much - so why is that? 

Cheers, Florian

P.S. @itasc guys: Pretty nice valueable work by all means!


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/robotics/attachments/20090926/083dff6d/attachment.htm