1

I'm trying to solve an IK problem using Jacobian matrices and their pseudo inverses to my human skeleton model, but I'm getting really awkward results. But the thing is, the answer is right! (that is, close enough to the target position & orientation) For example, here is one of the awkward results.

enter image description here

This is the IK result when I'm applying it to the entire upper part of his pelvis. The pelvis is the fixed point here. He's leaning back his upper body where he doesn't need to.

However, this is when I fix his shoulder location, and just apply IK to his arm to the very same target position. enter image description here

As you can see, you don't need to tilt the spine all the way to the back!

Is this a known problem of the Jacobian-IK method..? If so, is there any way that I can make this more human-like?


JIC, I calculated my Jacobian matrix like this enter image description here

Peter
  • 460
  • 6
  • 23
  • you need to add priorities to your bones ... that is very hard to do with your current algebraic approach that is why I recommended the CCD earlier... – Spektre Oct 30 '18 at 10:47
  • @Spektre I never thought this kind of problem would happen in the Jacobian approach : ( I will look up CCD then! – Peter Oct 30 '18 at 10:58
  • the catch with CCD is to order the iteration of bones ... and also reserve some bones for specific purpose (like just for turning head etc)... It can get really complicated mine CCD engine for stuff like this is ~40 KByte of C++ code not counting underlying libs like vector/matrix math ... – Spektre Oct 30 '18 at 11:00

0 Answers0