Animation Retargeting

I’ve skimmed many a document on animation retargeting, but I’ve never really seen anything that explains its strengths and limitations. Specifically, I’m mainly interested in how characters interact with each other and the environment.

When one character interacts with another. E.g., you have animations of a 6’ biped picking a 6’ biped up by the throat and slamming him against a wall. But say you want to animate the same thing with a 8’ biped and an 4’ biped instead. Will retargeting solve this, or is this something beyond the tech?

Another example, a 6’ biped doing the stereotypical showing the (6’ biped) alien how to shake hands scene (grabs wrist, pulls hand up, shakes it). Will retargeting solve this for 7’ and 4’ bipeds?

Or you have a 6’ biped walking up a set of stairs, will retargeting make the animations work right with a 4’ or 8’ biped?

Is it a matter of “the more you diverge from the original scale, the less effective retargeting becomes”? If so, what are the limits? I.e., at what point does retargeting start to break down?

Assume same skeletons are used between characters (e.g., 4’ guy or 7’ guy getting slammed have same skeletons, 4’ or 7’ guy doing the slamming have same skeletons).

Obviously there are other concerns outside the scope of what I’m asking (like physical limitations of the real world - kinda hard for a 4’ guy to pick a 7’ guy up by the neck; someone whose legs are too short to actually walk up a flight of stairs - we don’t necessarily want retargeting to work in that situations).

I can’t watch videos without jumping through onerous hoops, so if you want to use a video, make sure it’s awesome and really shows detailed info. :slight_smile:

ETA: the more I think about it, the less likely I think retargeting alone is sufficient. Sounds almost like something for a “smart” animation like Endorphin or Euphoria or whatever it was; the 7’ biped is going to have to bend over first before he can grab the 4’ biped’s hand, which seems well beyond the scope of retargeting.