Validation of Predictive Dynamics Tasks - Human Motion Simulation: Predictive Dynamics - page 204

CHAPTER
9

Validation of Predictive
Dynamics Tasks

With Contributions by Salam Rahmatalla, PhD, The University of Iowa

The biggest place I look for validation is from my mother. That's the little
girl in me that will never grow up.
Naomi Watts

9.1 Introduction
This chapter introduces a validation methodology to assess the predicted motion
of computer human models and provide feedback to model developers for evalua-
tion and refinement. Due to the varied strategies for (and the relative complexity
of) human motions under different dynamic tasks, the proposed validation meth-
odology is designed to be general and effective.
While objective and subjective statistical tools comprise the major components
of the proposed validation methodology, accurate collection of human data and
efficient handling of the data during post-processing operations are critical com-
ponents in this process. In order to demonstrate the validation method, we will
implement it in the two tasks detailed in Chapters 7 and 8.
Advances in measurement systems and motion capture technologies have
played a key role in the development of human modeling and simulation. Human
models have become more human-like with more sophisticated skeletons and pre-
dictive capabilities. As a result, many questions have been raised about the valid-
ity of these predictive models in representing human motions. One major question
is how to define the line between acceptable and unacceptable predicted motion;
there are many possible answers, depending on the required accuracy and applica-
tion. The objective of this chapter is to present effective validation methodologies
that provide tools to answer such questions.
Some approaches for validating predictive human-model motion have been
very helpful in the development and acceptance of the models ( Abdoli-Eramaki
et al., 2009 ; Blankevoort and Huiskes, 1996 ; Chaffin, 2002 , 2009; Dubowsky et al.,
2008; Karduna et al., 2001 ; Marras and Sommerich, 1991 ; Rabuffetti and Baroni,
1999; Robert et al., 2007 ). However, most of these approaches were designed for
specific usage or tasks, and they usually targeted a certain area of the human body
or used a low-fidelity model with a limited number of links and joints.