Professional Documents
Culture Documents
01404284.non Linear Pandy
01404284.non Linear Pandy
Abstract – Determining individual muscles forces from model actuated by multiple muscles within an optimal
human performance has greatly depended on the quality of tracking problem. In this approach, muscle excitation inputs
inverse dynamics solutions, as muscle force decomposition to a musculoskeletal model are adjusted via an optimization
remains the only feasible approach for determining muscle algorithm in order to minimize performance differences
forces non-invasively in human movement. However, between the model and the human. This approach, however,
legitimate questions about the accuracy of inverse dynamics requires thousands of forward simulations to be executed in
arise, with resultant torques/forces failing to drive a forward order to minimize the objective function, which is a total
model through the observations from which they were derived.
While optimization of forward dynamics to match
measure of tracking error. A comparison of a static
experimental data is considered more accurate, the simplicity decomposition approach based on inverse dynamics and a
and low computational costs of inverse methods are favored large-scale forward optimization for determining muscle
over the large computing requirements of optimization. In this forces in gait [3] concluded that the two methods produced
paper, an evolution in the inverse methods for computing similar results. Thus, based on inverse dynamics torques,
accurate and reliable torques is presented, whereby the researchers have begun to include muscle dynamics into the
relative speed of inverse dynamics is combined with the desired muscle force decomposition process [4], while others [5]
accuracy of forward dynamics. This method is based on employ EMG-driven muscle models to estimate muscle
developing a nonlinear tracker that determines the net muscle forces directly from EMG. In either case, torques computed
torques which accurately follow clinically observed kinematics
and ground reaction forces. The results show that the method
from inverse dynamics are used at some point to affirm the
is robust and can produce accurate estimates of the joint muscle forces obtained from the decomposition approach.
torques during movement. The method outlined here is a In studies reported by Anderson and Pandy [3], Thelen
necessary first step to solving the muscle force indeterminancy et al. [4], and Lloyd and Besier [5], the quality of the
problem more efficiently. inverse dynamics torques is not addressed. For example,
Anderson and Pandy [3] used the torques from their forward
Keywords – inverse dynamics, nonlinear control, computed optimization results to circumvent inverse dynamics, and
torques, feedback linearization, jumping thus their similarity conclusions are, in effect, based on
perfect inverse dynamics data. In reality, this is far from the
I. INTRODUCTION case, when one considers skin marker error, noise enhancing
effects of double differentiation, and the mechanical
Muscle forces are critical for quantifying muscle action differences between the human and the model. Inverse
in human performance and also for determining realistic dynamics is therefore inherently imprecise. Although low-
loading conditions for prosthetic devices such as hip or knee pass filtering and least squares estimations of segment
replacements. Although direct measurements of muscle rotations can mitigate some of the errors, torques computed
force from human performance would be ideal, this from inverse dynamics still may not reproduce measured
approach is obviously not feasible from an ethical joint movements and GRFs when applied in a forward
standpoint. Therefore, external observations such as marker simulation [6]. A simple reliability check would be to drive
positions, ground reaction forces (GRFs) and muscle EMG the model forward using inverse dynamics torques and to
are often used with computer models of the body to estimate compare the resultant kinematics and GRFs against
muscle forces non-invasively. The most common approach measured values. Unfortunately, such checks are rare even
is to use inverse dynamics to compute net joint torques and when the torques computed from inverse dynamics are the
to then use decomposition methods to estimate individual basis for estimating muscle forces.
muscle forces. Inverse dynamics requires the kinematics The purpose of this study is to develop a more robust
(including accelerations) and a link-segment model, from tracking method for computing net joint torques based on
which the net joint torques can be calculated. Specifically, inverse dynamics. A tracking method is required to ensure
force/torque balances are performed on the segments of a that the torques computed from inverse dynamics will drive
model from the ground up, where the GRFs and center-of- a model in a stable forward simulation and reproduce the
pressure are known from force-plate data. observed kinematics and ground reaction forces. The
In recent years, studies such as those by Neptune et al. method is demonstrated by computing the torques needed to
[1] and Zajac et al. [2] have employed a forward dynamics produce a maximum height jump. Even though the model
4634
those obtained from inverse dynamics. The computed
&y&q torques were virtually identical to the reference torques (Fig.
=v (6)
4) up until the time when the heel left the ground (0.21s);
y& s after heel lift-off, the hip and knee torques became erratic.
which is easily stabilized with linear feedback and the
appropriate gains on the tracking errors.
III. RESULTS
Fig. 3: Tracking (solid), inverse dynamics (dashed) and observed (shaded)
GRFs
4635
IV. DISCUSSION noise and sampling error for a highly dynamic task such as
jumping. Consequently, the tracking method determined
The nonlinear tracker performed well in reproducing the required input torques more accurately than
the performance of the model that matched the input conventional inverse dynamics. Controller design, however,
observations as illustrated in Figs. 2 and 3. An unexpected can be a challenging task, especially with ground contact
outcome was that the kinematics obtained from tracking models that include hard nonlinearities (i.e., discontinuous
more closely matched the reference kinematics than did the dynamics or gradients), which can inadvertently change the
observed kinematics. This occurred because both the joint controllability of the system and thus the requirements on
kinematics and the GRFs were tracked simultaneously. The feedback linearization. In such cases, it may be necessary to
feedback gains, therefore, ultimately dictate the trade-off in design multiple controllers for different phases of a single
tracking performance between the kinematics and the GRFs. task.
The fact that measurements of GRFs are much more The results of this study should prove useful for those
accurate than measurements of joint kinematics is fortuitous who wish to acquire quantitatively accurate net joint torques
because the performance of the tracking method can be from musculoskeletal models, especially for the purpose of
enhanced by having relatively higher gains on GRFs errors. decomposing joint torques into individual muscle forces.
In this way, the kinematics act to guide the model, whereas
the accuracy can be derived from the GRFs. Thus, the ACKNOWLEDGMENT
kinematics determined from the tracking method are more
reliable than the observed kinematics because the Supported in part by the National Science Foundation,
dependence of the kinematics on GRFs is enforced by virtue Engineering Research Centers Grant EEC-9876363. A. Seth
of the model. thanks the International Society of Biomechanics for the
Even though forward simulations using the torques support of a 2004-2005 ISB Dissertation Grant.
from inverse dynamics follow the general trends of the
kinematics (Fig. 2), the resulting GRFs are erroneous (Fig. REFERENCES
3). The torques from inverse dynamics over estimate the [1] Neptune, R. R., Kautz, S. A., and Zajac, F. E. (2001). Contributions
peak output of all the joints, especially the ankle before 0.1s. of the individual ankle plantar flexors to support, forward progression
In contrast the tracking torques reproduce the input torques and swing initiation during walking. Journal of Biomechanics,
well for most of the jump. When the heel leaves the ground 34:1387-98.
[2] Zajac, F. E., Neptune, R. R., and Kautz, S. A. (2002). Biomechanics
(0.21s), the hip and knee torques computed by the tracking and muscle coordination of human walking. part i: introduction to
method become erratic (Fig. 4). The reason is that the heel concepts, power transfer, dynamics and simulations. Gait and Posture,
contact point at that time is no longer governed by spring- 16:215-32
damper equations, but rather switches to a frictional limit [3] Anderson, F. C. and Pandy, M. G. (2001). Static and dynamic
optimization solutions are practically equivalent. Journal of
dependent on the vertical GRF. The hip and knee torques Biomechanics, 34:153-61.
are unable to fully recover from this transition for a few [4] Thelen, D. G., Anderson, F. C., and Delp, S. L. (2003). Generating
reasons. First, the ankle torque dominates the interaction dynamic simulations of movement using computed muscle control.
with ground just prior to takeoff [10], and so the input GRFs Journal of Biomechanics, 36:321-328.
[5] Lloyd, D. G. and Besier, T. F. (2003). An emg-driven
predominantly influences the ankle control, which also musculoskeletal model to estimate muscle forces and knee joint
explains the accurate ankle torque throughout the tracking. moments in vivo. Journal of Biomechanics, 36:765-776.
The tracker, therefore, must depend more on the joint [6] Risher, D. W., Shutte, L. M., and Runge, C. F. (1997). The use of
kinematics for determining the knee and hip torques. inverse dynamics solutions in direct dynamics simulations. Journal of
Biomechanical Engineering, 119:417-422.
Second, the kinematics, which are typically less accurate, [7] Anderson, F. C. and Pandy, M. G. (1999). A dynamic optimization
are even more error prone closer to takeoff due to position solution for vertical jumping in three dimensions. Computer Methods
sampling when the jumper is moving at its peak velocity. in Biomechanics and Biomedical Engineering, 2:201-231.
Thirdly, the heel leaving the ground essentially changes the [8] Runge, C. F., Zajac, F. E., Allum, J. H. J., Risher, D. W., Bryson, A.
E., and Honegger, F. (1995). Estimating net joint torques from
controllability of the system. Specifically, as the foot kinesiological data using optimal linear system theory. IEEE
translates upward the contact damping diminishes (at 5mm Transactions on Biomedical Engineering, 42(12):1158-1164.
above the ground) thereby rendering the controller less [9] Seth, A., McPhee, J. J., and Pandy, M. G. (2003). Multi-joint
effective, because it is through the velocity term (4) that the coordination of vertical arm movement. Journal of Applied Bionics
and Biomechanics, 1(1):45-56.
feedback linearization was obtained. [10] Bobbert, M. F. and van Ingen-Schenau, G. J. (1988). Coordination in
vertical jumping. Journal of Biomechanics, 21(3):249-262.
V. CONCLUSION [11] Pandy, M. G., Zajac, F. E., Sim, E., and Levine, W. S. (1990). An
optimal control model for maximum-height human jumping. Journal
of Biomechanics, 23(12):1185-1198.
Overall, simultaneous tracking of GRFs and joint
kinematics achieved accurate results despite the addition of
4636