This paper tries to address the challenges that occur in doing
online puppetry where we don't have knowledge of future frames.
There are two major challenges, one is the need for motion
retargeting since the proportions of the performer and the
character is different, and the other is the noises that are
more severe on the magnetic type of sensors that are more
effective for online applications.
The paper suggests solutions for both of these problems.
For noise, the paper suggest using Kalman filters to predict
future poses and smoothen out the noise.
For retargetting, the paper suggests ways to select between
if the angle of the joint or the end-effector position is more
important. The paper suggests that
1. Root position isn't very important
2. End-effector interacting with environment is important
3. End-effector about to interact with environment is important
4. If end-effector finished interacting with environment it is less important
5. Otherwise joint angle is more important