Validation of Predictive Dynamics Tasks - Human Motion Simulation: Predictive Dynamics - page 229

The results of the second qualitative benchmark test have shown some similar-
ity between the shapes of the predicted and human lifting determinants as shown
in Figure 9.16A and Figure 9.16B . However, the model showed poor performance
in following the shape of the ankle and shoulder flexion during the lifting cycle.
The predicted model showed some potential for capturing the coupling between
the lifting determinants as shown in Figure 9.16B , with some poor performance
in capturing the coupling between the shoulder and the ankle. The importance of
this benchmark test is to detect major discrepancies, if there are any.
Additionally, this test can identify any abnormal characteristics in the shape of
the determinants' time history during the lifting cycle.
In terms of the quantitative comparison in the third benchmark test, the pre-
dicted lifting motion showed a reasonable correlation with the experimental data
( Figure 9.18 ), where all determinants, except the shoulder flexion-extension and
the ankle flexion angles, stayed inside the interval of confidence at all times and,
notably, followed the mean of the subjects. The shoulder flexion-extension angle
showed acceptable correlation during the first half of the lifting cycle, but poorly
represents human characteristics during the second half of the motion cycle.
These discrepancies could be related to model accuracy, in that the experimental
data was collected from a real human with muscle and skin movement, while the
model is based on a rigid-body dynamics assumption.
The discrepancies may also be due in part to the absence of necessary con-
straints on the complex shoulder motion in the model and in part to the difficul-
ties associated with computing accurate shoulder joints from the experimental
data. The discrepancies could also be attributed to other parameters, such as the
complex interaction of the shoulder-clavicle complex motion and the motion of
the wrists when the box is extended above chest level while keeping the feet on
the floor. Ankle flexion showed behaviors similar to those of shoulder flexion.
Unlike other human motion tasks that require a single common strategy, such
as walking, the box-lifting task can be conducted using different strategies; there-
fore, the whole motion looks natural, but the determinants' shapes could be
affected by the coupling and the speed associated with the strategy being used.
As shown in Figure 9.21 , the model inadequately captured the early stages of the
lifting process, which could involve different speeds and different lifting strate-
gies; still, the model succeeded in capturing the following steady-state phase for
both legs. It should be noted here that the GRF for box lifting were based on one
subject and may be not sufficient for use in model kinetics validation due to the
uncertainty in the lifting strategy and the possibility of the subjects leaning lat-
erally. Therefore, the reaction forces at the hands should also be measured and
used in the comparison. However, in this work, we considered symmetrical lifting
and assumed equal distribution of the forces on the hands and legs.
For the key frames in the fourth and last quantitative benchmark test,
Figure 9.22 demonstrates the correlation between the experimental and simulated
determinants with R 2 extended from 0.72 to 0.98 for most determinants, with the
exception of the shoulder flexion.