Validation of Predictive Dynamics Tasks - Human Motion Simulation: Predictive Dynamics - page 220

Selected key-frames (walking)
100

80

Knee flexion:
y= 1.0827x + 11.388
R

2

-0.9214

60

40

Pelvic rotation:
y= 0.8799x - 1.2274
R
2

Hip flexion:
y= 1.1032x + 0.2335
R

= 0.8355

2

-0.9779

20

0

-80 -60 -40 -20 0 20 40 60 80 100

-20

Pelvic displacement:
y= 0.7192x + 0.193
R

Pelvic tilt:
y= 1.9916x + 0.0943
R

2

= 0.8255

2

= 0.6291

Hip_Flexion
Knee_Flexion
Ankle_Flexion
Pelvic_Tilt
Pelvic_Rotation
Pelvic_Displacement

-40

Ankle flexion:
y= 0.974x - 2.5383
R

2

= 0.5324

-60

-80

Experiment

FIGURE 9.13
R 2 plot for the six walking determinants for the selected key frames (bottom). The vertical
axis (in degrees) stands for simulation data (Santos s ), and the horizontal axis (in degrees)
corresponds to the average experimental data.

For the key frames, Figure 9.13 demonstrates a good correlation for the hip
flexion, as indicated by the value of R 2 , which exceeds 0.9. Similar behaviors
with less correlation can be seen for other determinants. However, the ankle
dorsi-flexion has shown relatively poor correlation, with R 2 less than 0.53. For
the GRF, Figure 9.10 demonstrates a good agreement between the two curves in
the vertical and the fore-aft directions. This gives the model significant value and
potential for calculating realistic values for the joint torque of the different

human's joints.
The determinant's velocity for the predicted model has shown similar charac-
teristics to those of the subject population. However, the predicted model could
not capture the higher-frequency components, which are represented by the local
fluctuation in the velocity curves. Normally, natural human motion is relatively
smooth, and the local fluctuations, shown in Figure 9.9A , are anomalies that may
represent some type of numerical noise as a result of the finite differences calcu-
lation, or because of local movement of reflective markers due to skin motion
( Garling et al., 2007; Lucchetti et al., 1998 ). A low-pass filter may be used to
smooth out the experimental data. What has been shown and applied to velocity
seems to be applicable to the acceleration, as shown in Figure 9.9B .