Lecture 5

Finished covering IK, started motion editing.

Thoughts

  • IK: same thoughts as previous lecture
  • FABRIK was pretty impressive, at least in the unconstrained case. I don’t really know if there’s a rigorous theoretical basis for it, but the simple algorithm gave rather natural behavior.
  • I thought the Jacobian techniques were generally more suitable for actual motion planning, since they always describe the continuous motion. However, it seemed the more straightforward (though iterative) solvers worked for this well enough, since each iteration gives some step in the right direction. Or, if we’re given some pre-determined path for the end effector to follow, it can just run a solve to get the change in angles for each small step along the trajectory. Both options do have the potential to require some discontinuous motion, though, and still don’t seem quite as useful (or as easy to maintain constraints) if you want to generate the motion plan from scratch.
  • Mesh IK and associated automatic rigging/posing systems were quite cool, although we didn’t get too much into their technical details.
  • I look forward to talking more about learned motion synthesis with mocap data, etc.
Written on February 17, 2021