Posture Prediction and Optimization - Human Motion Simulation: Predictive Dynamics - page 45

is the ability to predict postures that already have been recorded for the exact
anthropometric model. Extrapolating postures and variations thereof is extremely
difficult and highly inaccurate. However, where this method completely fails is in
predicting motion, which is the ultimate objective of posture prediction. If it is to
be expanded to predict motion, the variability and many parameters associated
with motion, including dynamic effects and inertia, are not only difficult to mea-
sure but impossible to correlate. The method requires an exhaustive and often
very costly experimental setup involving thousands of subjects to generate a mod-
est model for a small population.

3.5 Optimization-based posture prediction
This chapter introduces a framework and associated algorithm for predicting pos-
tures that are based on an individual task. In order to better understand the moti-
vation behind cost functions, consider the case of a driver, in a vehicle, who is
about to reach for a radio control button on the dashboard. It is believed that the
driver will reach directly to the button while exerting minimum effort and perhaps
expending minimum energy. However, the same driver when negotiating a curve
will have to place their hand on the steering wheel in such a way to be able to
exert the necessary force needed to turn the wheel. As a result, involuntarily, the
driver will select a posture that maximizes force at the hand, minimizes the torque
at each joint, minimizes energy, and minimizes effort needed to accomplish this
task ( Figure 3.2 ).
Therefore, our underlying plot is that each task is driven by the optimization
of one or more cost functions, leading the person to posture their body in the
most natural manner. Note that simple logic has been implemented in the
Processor module to correlate between the task and the cost functions.
What we have proposed through this simple example is that humans assume
different postures for different tasks. Note that, when coupled with the DH

Cost Functions

Task
Turn steering wheel
Turn headlights on/off
Shift gear
Adjust mirror
...

Effort
Energy
Dexterity
Reach
Discomfort
Force at fingertip
Torque at joint
Stress

min
min

Processor

min

FIGURE 3.2
The task-based approach to selecting cost functions.