Fig 1.
Relevant curves for our Hill-type muscle and activation nonlinearization model.
Left: Normalized active, passive, and total force-length curves. Middle: Normalized force-velocity curve. Right: Neural-to-muscle activation nonlinearization curves for minimum nonlinearization (curve parameter = 0) and maximum nonlinearization (curve parameter = 0.35).
Table 1.
List of muscles in the model, which DOF each muscle actuates, and source of each muscle’s excitation signal.
Fig 2.
Flowchart of EMG-driven model calibration process for walking.
The goal is to find model parameter values (i.e., activation parameters, surrogate geometry parameters, and muscle-tendon parameters) such that experimental processed EMG data and joint kinematics can be input to the model and lower extremity joint moments that closely match experimental joint moments are output from the model. Blue lines indicate model parameter values changed by the optimization process.
Fig 3.
Average joint moment predictions for walking at 0.5 m/s when calibrating using all five walking speeds.
NGA stands for no geometric adjustments and WGA stands for with geometric adjustments. Average experimental values with gray bands specifying +/- 1 standard deviation are shown for visualization purposes and were calculated at each time point after each gait cycle was resampled to 101 points.
Fig 4.
Average joint moment predictions for walking at 0.8 m/s when calibrating using all five walking speeds.
NGA stands for no geometric adjustments and WGA stands for with geometric adjustments. Average experimental values with gray bands specifying +/- 1 standard deviation are shown for visualization purposes and were calculated at each time point after each gait cycle was resampled to 101 points.
Table 2.
Mean MAE values for testing trials using EMG-driven models calibrated at all available walking speeds without (NGA) and with (WGA) geometric adjustments.
The percent change in MAE when geometric adjustments were added is also reported, with the standard deviation of MAE between trials shown in parenthesis.
Fig 5.
Average joint moment predictions for walking at 0.5 m/s when calibrating using only the three slowest walking speeds.
NGA stands for no geometric adjustments and WGA stands for with geometric adjustments. Average experimental values with gray bands specifying +/- 1 standard deviation are shown for visualization purposes and were calculated at each time point after each gait cycle was resampled to 101 points.
Fig 6.
Average joint moment predictions for walking at 0.8 m/s when calibrating using only the three slowest walking speeds.
NGA stands for no geometric adjustments and WGA stands for with geometric adjustments. Average experimental values with gray bands specifying +/- 1 standard deviation are shown for visualization purposes and were calculated at each time point after each gait cycle was resampled to 101 points.
Table 3.
Mean MAE values for testing trials using EMG-driven models calibrated at 0.4, 0.5, and 0.6 m/s walking speeds without (NGA) and with (WGA) geometric adjustments.
The percent change in MAE when geometric adjustments were added is also reported, with the standard deviation of MAE between trials shown in parenthesis. The bold row headers indicate the gait speeds being predicted that were not included in model calibration.
Fig 7.
Passive joint moment matching.
Passive moments predicted by our EMG-driven models calibrated using all walking speeds (dashed lines) compared to published passive moments (solid lines) for the WGA and NGA models.
Table 4.
Comparison of moment error values reported in the literature with moment error values reported in this study.
Other EMG-driven studies not indicated [12,16–19] have prediction errors greater than those listed in this table or use a variety of activities for calibration and/or testing and are therefore disqualified from comparison. For the knee and ankle joints, the studies shown calibrate and test their models using only gait data. Sartori et al. 2014 was the only available EMG-driven model of the hip, and was calibrated using a variety of activities.