Posture Prediction and Optimization - Human Motion Simulation: Predictive Dynamics - page 44

for a different set of cost functions. It is expected that this aspect of selecting cost
functions will continue to be an area of research of great interest as it produces
behaviors that vary for performing the same task with different cost functions.
The main new idea here is that humans execute procedures while minimizing
or maximizing cost functions. By selecting a combination of various cost func-
tions, a behavior is induced and a new motion profile is calculated. This concept
will be explained further in this chapter.
3.4 Posture prediction versus inverse kinematics
Because of the apparent mathematical complexity of the problem of posture pre-
diction and because the human body has many more than six DOF, methods for
predicting postures are quite involved. Posture prediction, at least on the surface,
is equivalent to what is called inverse kinematics in the field of mechanisms and
robotics (known as IK in the gaming and animation industries). In this section we
comment about their benefits and shortcomings.
3.4.1 Analytical and geometric IK methods
Analytical and geometric IK methods are typically associated with robotic manip-
ulator arms that have no more than 6 DOF and where closed-form or numerical
solutions are possible. Larger than 6-DOF systems become very redundant and
require very complex numerical algorithms. Analytical or geometric IK methods
for human posture prediction are almost impossible because of the following
reasons:
1. The large number of DOF associated with the human model leads to severe
difficulties in calculating solutions. It is almost impossible to use geometric
methods and very difficult to use numerical methods, as identifying and
finding all solutions is a difficult mathematical problem.
2. The need to calculate realistic postures. While some of the analytical methods
may yield solutions, a choice of “the best” solution that looks most natural is
difficult to achieve using analytical IK methods.
The main benefit of analytical and geometric methods is that, if determined,
they are determined very quickly. Typically, analytical methods, even though
they are numerical in nature, are computationally efficient.
3.4.2 Empirically-based posture prediction
This method is based on gathering a great deal of data from actual subjects while
they are performing various postures. Anthropometry and posture data are cap-
tured and recorded. Statistical models, typically nonlinear regression, are then
developed and used in posture prediction algorithms. The benefit of this method