ARTICLE

Forward Dynamics Simulations Provide Insight

Into Muscle Mechanical Work During Human
Locomotion
Richard R. Neptune,1 Craig P. McGowan,1 and Steven A. Kautz 2,3,4
1
Department of Mechanical Engineering, The University of Texas at Austin, Austin, TX; and 2Brain Rehabilitation
Research Center, Malcolm Randall VA Medical Center; and 3Department of Physical Therapy and 4Brooks
Center for Rehabilitation Studies, University of Florida, Gainesville, FL

NEPTUNE, R.R., C.P. MCGOWAN, and S.A. KAUTZ. Forward dynamics simulations provide insight into muscle mechanical
work during human locomotion. Exerc. Sport Sci. Rev., Vol. 37, No. 4, pp. 203Y210, 2009. Complex musculoskeletal models and
computer simulations can provide critical insight into muscle mechanical work output during locomotion. Simulations provide both a
consistent mechanical solution that can be interrogated at multiple levels (muscle fiber, musculotendon, net joint moment, and
whole-body work) and an ideal framework to identify limitations with different estimates of muscle work and the resulting implications
for metabolic cost and efficiency. Key Words: muscle power, energy, metabolic cost, efficiency, mechanics

INTRODUCTION Of particular interest to analyses of human movement is

muscle mechanical work, as it is often used to compare
Complex musculoskeletal models and computer simula- estimates of the work required to perform locomotor tasks
tions are becoming an integral part of analyzing human such as walking and running; understand how energy is gen-
movement over a wide range of topics, from understanding erated, absorbed, and/or transferred among the body seg-
fundamental muscle coordination principles (e.g., (35)) to ments; identify impairments associated with neurological
identifying potential injury mechanisms (e.g., (26)). Muscle- deficits; and estimate metabolic cost and efficiency. We have
actuated forward dynamics simulations are particularly pow- previously used simulation analyses to gain insight into the
erful because they allow for the identification of causal mechanical work requirements of human locomotion and
relationships between the neural control inputs, muscle force elucidate limitations associated with experimentally derived
and power output, and the specific task performance (see measures of muscle work. Our central premise is that exter-
(35) for review). Understanding these relationships without nal measures of mechanical work cannot accurately quantify
simulation analyses is challenging because of the highly work performed by muscles and, as a consequence, may bias
complex and nonlinear nature of the musculoskeletal system, estimates of efficiency during locomotion. Accordingly, de-
with its corresponding dynamic coupling that causes muscles tailed forward dynamics simulations and metabolic energy
to influence the system in counterintuitive ways. Simulations cost models derived from experimental data will be critical
also allow estimation of quantities that are difficult or im- to understanding the mechanics and energetics of human
possible to measure in vivo, such as joint and tissue loading, locomotion.
muscle fiber and/or tendon force and power, and elastic
energy storage and return in tendons.
RELATIONSHIPS BETWEEN INTERNAL, EXTERNAL,
JOINT, AND MUSCLE WORK

Address for correspondence: Richard R. Neptune, Ph.D., Department of Mechanical A number of methods have been used over the years to
Engineering, The University of Texas at Austin, 1 University Station C2200, Austin, estimate the total muscle mechanical work with internal,
TX 78712 (E-mail: rneptune@[Link]). external, and/or joint workYbased quantities. Internal work is
Accepted for publication: June 9, 2009.
Associate Editor: Daniel P. Ferris, Ph.D. defined as the sum of the absolute changes in body segment
kinetic and potential energy and is considered to be the work
0091-6331/3704/203Y210
Exercise and Sport Sciences Reviews
necessary to move the body segments relative to the body’s
Copyright * 2009 by the American College of Sports Medicine center of mass ((COM) e.g., (33)) or with respect to an

203

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
inertial reference frame (e.g., (34)). External work is defined passive structures) and the level of cocontraction can be
as the mechanical work done on the body’s COM (e.g., precisely quantified. We used simulation analyses of pedaling
computed as the time integral of the product of the measured to demonstrate how the work done by individual muscles to
ground reaction force and body COM velocity during accelerate the leg segments ends up as external work at the
walking or running (33)) or an external load (e.g., during pedal (13). Muscles accelerate the leg segments early in the
pedaling (32)). Joint work, which is generally considered to downstroke and that energy is recovered later in the down-
be a more accurate estimate of musculotendon work than stroke as the legs do mechanical work on the external load,
external or internal work, is computed as the time integral of which results in deceleration of the leg segments. As a result,
net joint power calculated using standard inverse dynamics significant energy is transferred from the legs to the crank
techniques. Although these measures have been used in a without negative work being done by the muscles to de-
wide range of studies to estimate muscle work, simulation celerate the leg segments. Thus, internal work, which at-
studies have been useful in quantifying the acknowledged tributes all reductions in system energy to muscles, does not
limitations with these methods (i.e., lack of independence provide a valid measure of the energetic cost of moving the
between internal and external work, inability to account for legs in pedaling. These results provided theoretical validation
cocontraction and elastic energy storage and return, etc.). for our experimental work that showed that large changes in
Simulations provide an ideal framework to make this internal work were not associated with similar changes in
assessment because they provide an example performance in joint work during pedaling (12). Similarly, we used other
which every source of mechanical work associated with a pedaling simulations to assess the ability of various measures
given locomotor task (e.g., from muscle fibers, tendons, and of mechanical energy expenditure (MEE) to estimate muscle

Figure 1. The musculoskeletal model and optimization framework used to generate forward dynamics simulations. The model consisted of a trunk
(head, arms, torso, and pelvis) and left and right legs (femur, tibia, patella, rear foot, midfoot, and toes). Only the 13 muscle groups for the right leg are
shown, which included the anterior and posterior portion of gluteus medius (GMED); iliacus, psoas (IL); rectus femoris (RF); 3-component vastus (VAS);
tibialis anterior, peroneus tertius (TA); peroneus longus, peroneus brevis (PER); flexor hallucis longus, flexor digitorum longus (FLXDG); extensor hallucis
longus, extensor digitorum longus (EXTDG); soleus, tibialis posterior (SOL); medial and lateral gastrocnemius (GAS); biceps femoris short head (BFsh);
medial hamstrings, biceps femoris long head (HAM); and gluteus maximus, adductor magnus (GMAX). The Optimization Algorithm fine-tunes the muscle
excitation patterns for each muscle group to produce a well-coordinated walking pattern that emulates the human subject data by minimizing cost
function J. Y is the experimental data; Ŷ is the corresponding simulation data. The excitation patterns are defined by muscle EMG or a rectangular pattern
if no EMG data are available. As seen in the Optimal Excitation Patterns, the resulting excitation patterns agree well with the human subject EMG data.
For comparison purposes, the experimental EMG patterns (group mean T SD) were normalized to the peak simulation excitation magnitude. The
excitation patterns for the small muscles that primarily control the foot (PER, FLXDG, EXTDG) are omitted in the Optimal Excitation Patterns comparison.
The Compare Output to Experimental Data shows how well the simulation emulates the experimental hip, knee, and ankle joint angles, vertical and
horizontal GRF, and joint moments over the gait cycle (i.e., from right heel-strike to right heel-strike). [Adapted from Neptune RR, Clark DJ, Kautz SA.
Modular control of human walking: A simulation study. J. Biomech. 2009; 42(9):1282Y7. Copyright * 2009 Elsevier. Used with permission.]

204 Exercise and Sport Sciences Reviews [Link]

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
mechanical work (17). Two distinct pedaling simulations integral of the muscle joint power at each joint spanned by a
were generated at the same pedaling rate and power output, muscle, which is equivalent to joint power in the absence of
with one optimized to reduce MEE by minimizing eccentric passive joint power) and total musculotendon work (i.e., the
muscle contractions. Comparisons were made between muscle time integral of the musculotendon power, which is the
fiber work and MEE, which was quantified by three methods: product of corresponding musculotendon force and velocity
1) internal work, 2) net joint work, and 3) intercompensated vectors). Cocontraction is inevitable in most human move-
joint work (i.e., negative work from one joint was transferred ment tasks due to joint stability needs, movement control
to the adjacent joint, where energy is being generated [posi- complexity, synergistic muscle activity, and the influence of
tive work] via biarticular muscles). The three MEE measures activation and deactivation dynamics. Note that in contrast
greatly underestimated the muscle fiber work with errors to our findings for total joint work, we did find that when the
reaching as high as 40% (17). Errors in the MEE methods net passive joint work over the gait cycle is negligible (e.g.,
were mainly attributed to their inability to account for co- during slow walking), net joint work (i.e., the time integral of
contractions of antagonistic muscle groups. The simulations net joint power) was comparable to the net musculotendon
showed that muscles can absorb considerable power even work (and net muscle fiber work because net tendon work
when the net joint power is positive, which occurred in ped- is zero over a complete gait cycle). Thus, during walking
aling primarily due to one muscle generating positive work conditions when passive joint work is negligible, net joint
during activation while the antagonist muscle continues to work provides a reasonable estimate of net muscle work.
generate force (and thus negative work) during deactivation. However, net muscle work is of limited use since it provides
Recently, we used simulations to investigate the ability of neither insight into total mechanical work nor metabolic cost
internal, external, and joint work to estimate muscle mechan- or efficiency.
ical work in human walking (21). Unlike in pedaling, where The simulation analysis also showed that neither total
the pedal reaction force rotates the crank, walking has no external nor total internal work (nor their sum) provided a
associated displacement at the end point due to the external reasonable estimate of total musculotendon work, which is
reaction force. Thus, external work has been defined as the consistent with the pedaling studies previously noted
work done by the ground reaction force on the body’s (12,17). External work in particular is often used to infer
COM. However, similar to pedaling, there are regions of co- muscle work and metabolic cost in walking (e.g., (6)).
contraction that occur during the gait cycle, and unlike ped- However, external power is a net measure that cannot parti-
aling, energy is likely dissipated when the feet collide with tion out the amount of positive and negative muscle work
the ground during walking. Thus, it was unclear if internal, that is performed during a given locomotor task and thus
external, or joint work could provide a more reasonable es- is not well suited for predicting muscle work or metabolic
timate of musculotendon and/or muscle fiber work during cost. Support for these limitations predicted by simulation
walking than they do during pedaling or if these methods analyses is provided in a recent study of amputee walking
would be subject to the same limitations. that showed that despite a 26% increase in metabolic cost
To address these questions regarding the use of internal, compared with nonamputees walking at the same speed
external, and joint workYbased quantities to estimate muscle (1.3 mIsj1), there was no difference in total absolute ex-
work, a muscle-actuated forward dynamics walking simula- ternal work, and there was no correlation between metabolic
tion was generated that emulated subjects walking at 1.5 mIsj1 and total external mechanical power (10). Thus, simulations
(Fig. 1). The simulation was generated using SIMM/Dy- were useful in showing that joint work is limited in its ability
namics Pipeline (MusculoGraphics, Inc., Santa Rosa, CA) to estimate musculotendon work, and that external and/or
and SD/FAST (Parametric Technology Corporation, Need- internal work does not provide a reasonable estimate of
ham, MA). Dynamic optimization was used to fine-tune the musculotendon work.
muscle excitation patterns with a cost function that mini-
mized the difference between the simulation and experimen-
tal kinematics (i.e., the time history of the trunk trajectory
and hip, knee, and ankle angles) and ground reaction force MUSCLE WORK REQUIREMENTS DURING WALKING
(GRF) data such that the simulation emulated well the group
averaged subject data. Constraints were placed on the Inverted pendulum models of walking predict that little
excitation magnitude and timing in the optimization muscle work is required for the exchange of body potential
algorithm to ensure that the muscles generated force in the and kinetic energy in single-limb support. Based in part on
appropriate region of the gait cycle. From the simulation, external power analyses, previous studies have suggested that
total joint work (i.e., the time integral of absolute value of the primary mechanical energetic cost of walking is associ-
joint power over a full gait cycle) was found to underesti- ated with the requirement to redirect the motion of the
mate total musculotendon work caused primarily by agonist- COM in double support during the step-to-step transition
antagonist cocontractions (21). This occurred despite the (6,14), which is most efficiently done by the ankle plantar
inclusion of joint work by passive structures and biarticular flexors (14). However, these simple inverted pendulum
muscle work, each of which acted to decrease the under- models without articulated joints do not capture the multi-
estimation and cannot normally be accounted for in an in- muscle multisegmental properties of walking, coexcitation of
verse dynamicsYbased calculation of total joint work. Muscle muscles to coordinate segmental power flow, or simultaneous
cocontractions can easily be determined in the model as the production of positive and negative muscle work. We used
difference between total muscle joint work (i.e., the time a muscle-actuated forward dynamics simulation of walking

Volume 37 c
Number 4 c
October 2009 Muscle Work in Human Locomotion 205

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
Figure 2. Comparison between experimental positive (Pos), negative (Neg), net (Net), and total (Total) joint work (A), simulation joint work (B), and
corresponding simulation musculotendon work (C) in each region defined by the positive/negative crossing points of the external power trajectory
(see external power in Fig. 1). Also shown in (C) is the simulation external work (External). The simulations showed that the musculotendon work is
significant in Region 2, which is necessary to decelerate hip and knee flexion, provide lower-limb stability, and accelerate the body center-of-mass
vertically to provide body support.

to address two important questions inspired by step-to-step to accelerate the hip and knee into extension and the COM
transition theory (18): 1) can the potential and kinetic en- vertically, thus raising the COM and providing body support
ergy of the body be exchanged with little muscle work; and (18). Without this muscle force (work), the joints would
2) does the net work output and the mechanical energetic collapse rather than extend, and the trunk would not rise.
cost for muscles occur primarily in double support. Note that These simulation results are consistent with Sawicki and
we have generated improved simulations since our original Ferris (22) who used metabolic measurements and powered
publication that more closely replicates experimental human ankle exoskeletons to replace ankle plantar flexor power
subject data (Fig. 2), and the conclusions remain robust. during push-off in double support (Region 4). They found
Contrary to the predictions of step-to-step transition theory, that at the preferred speed/step length combination (1.25
we found that significant muscle work occurs throughout the mIsj1), elastic energy stored in the Achilles tendon ac-
gait cycle and particularly in early single-limb support (Fig. 2, counted for 59% of the total ankle musculotendon positive
Region 2). In this region, cocontraction occurs among the work, and that the positive work generated by the ankle
hip and knee flexors and extensors to control hip and knee muscles in late stance only accounted for 18% of the total
flexion (e.g., knee flexion peaks near 30 degrees) and net metabolic power. They concluded that the metabolic
provides lower-limb stability. Muscle work is then needed cost of walking is most likely dominated by positive muscle

206 Exercise and Sport Sciences Reviews [Link]

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
Figure 3. Total muscle fiber and tendon work across increasing walking speeds (net musculotendon work (MTnet), positive tendon work ((TenPos)
elastic energy recovered from the tendon), negative fiber work (FibNeg), and positive fiber work (FibPos). The percent difference in muscle work between
stance and swing decreases as walking speed increases. Note that the percent of elastic energy recovery in the tendons (TenPos/(FibPos + TenPos)) was
maximized at 30% when walking at 1.2 mIsj1, which is near the preferred walking speed. [Adapted from Neptune RR, Sasaki K, Kautz SA. The effect of
walking speed on muscle function and mechanical energetics. Gait Posture 2008; 28(1):135Y43. Copyright * 2008 Elsevier. Used with permission.]

work at the proximal joints (i.e., hip and knee) that swing phase decreased. As a result, the positive fiber work
accounts for the remaining 82% of the total net metabolic performed during swing increased from 25% to 29% of the
power. The results of Sawicki and Ferris (22) support the total fiber work (stance + swing) with increasing walking
simulation results and further highlight the inherent limi- speed (Fig. 3). These results were consistent with the es-
tations of studies that rely on external power to infer muscle timate of Doke et al. (5) that moving the legs represents
mechanical work or metabolic cost. nearly one third of the total energetic cost of walking at
1.3 mIsj1, and Umberger (27), who used a metabolic cost
model with simulation analyses to estimate that 29% of the
THE EFFECT OF WALKING SPEED ON MUSCLE total metabolic cost of walking occurs during swing. In-
MECHANICAL WORK terestingly, the simulations also showed that walking near
self-selected speeds (1.2 mIsj1) improved the elastic energy
Simulations also have provided novel insights into under- storage and recovery in the uniarticular ankle plantar flexor
standing the relationship between individual muscle function tendons and reduced negative fiber work relative to faster or
and mechanical work as walking speed increases. During slower speeds (Fig. 3), which has implications for under-
normal walking, we often need to modulate speed over a standing preferred walking speed selection. Thus, simulation
large range; however, there is poor understanding of how the analyses proved useful in gaining insight into the neuromotor
neuromotor patterns adapt to the changing biomechanical mechanisms underlying speed regulation in walking, and the
demands of increased speed. Thus, we used muscle-actuated results provide a basis for comparison with pathological
forward dynamics simulations to identify functional and populations.
energetic adaptations in individual muscles to walk at faster
steady-state speeds (16). We expected the largest adaptations
to occur in stance where muscles provide body support and RELATING MUSCLE MECHANICAL WORK TO
forward propulsion while recognizing that increases in the METABOLIC COST
mechanical work of the swing phase muscles would likely be
needed to account for the increased importance of accelerat- An important area for future research will be to better
ing and decelerating the leg as walking speed increased. The understand the efficiency of musculoskeletal systems during
simulations showed that the relative contributions of dif- natural movements. Traditionally, mechanical efficiency has
ferent muscle groups to trunk support, forward propulsion, been expressed as the ratio of positive work (e.g., joint work,
or leg swing (i.e., muscle function) remained invariant, external work, or muscle work) to metabolic cost, although
although the magnitude of each contribution systematically some authors have suggested that negative work also should
increased with speed (16). Trunk support (vertical acceler- be included in the denominator of the efficiency calculation
ation) was primarily provided by the hip and knee extensors (efficiency = W+/(E + Wj), where W+ is the positive work,
in early stance and the plantar flexors in late stance, whereas Wj is the negative work, and E is the metabolic energy
trunk propulsion (horizontal acceleration) was primarily pro- expenditure) because negative work has the potential to be
vided by the soleus and rectus femoris in late stance. The returned as positive work later in the gait cycle (19). Using
simulations also highlighted the importance of initiating and the latter approach, Umberger and Martin (30) examined
controlling leg swing because there was a dramatic increase the influence of walking stride rate on efficiency using joint
at the higher walking speeds in iliopsoas muscle work to ac- work. Their results showed that when walking at 1.3 mIsj1,
celerate the leg in preswing and early swing, which coincided subjects had an efficiency of 0.38. To compare the results
with an increase in the biarticular hamstring muscle work to from our walking simulations with the experimentally deter-
decelerate the leg in late swing. At higher walking speeds, mined efficiency, we used the same equation and metabolic
the difference in positive fiber work between the stance and cost as Umberger and Martin (30) to calculate mechanical

Volume 37 c
Number 4 c
October 2009 Muscle Work in Human Locomotion 207

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
Figure 4. A schematic of a general muscle metabolic cost model implemented in a forward dynamics simulation framework. Activation heat rate (Ȧ)
and maintenance heat rate (Ṁ) are commonly grouped together and depend on the muscle activation and the percentage of fiber types for each muscle.
Basal heat rate (Ḃ) is based on the basal metabolic rate of muscle and is a function of muscle mass. Shortening heat rate (Ṡ) is a function of muscle
activation, muscle fiber type, and rate of muscle shortening or lengthening. Mechanical work rate (Ẇ ) is the product of muscle force and rate of muscle
length change. The total metabolic rate for each muscle is the sum of the rates of heat production and mechanical work.

efficiency from our simulation joint work (21). The results Schenau et al. (31) and the subsequent responses for a
showed that the simulation had similar efficiency (È0.40) as detailed discussion).
the experimental subjects. However, using simulations, we
were also able to examine efficiency at the muscle fiber level
(positive fiber work/metabolic cost) and found an efficiency METABOLIC COST MODELS
of 0.59. Note that we did not need to include negative fiber
work in the denominator as the theoretical upper limit for To further understand the relationships between muscle
how much elastic energy could be returned as positive work mechanical work and efficiency, detailed models of muscle
later in the gait cycle because we could extract the fiber metabolic cost are needed. There are multiple approaches to
work directly from the simulation. Sawicki and Ferris (22) modeling the energetics of muscle contractions. Since A.F.
computed ankle joint musculotendon efficiency across speeds Huxley’s original sliding-filament model, a number of authors
and found values near 0.61 at the preferred speed/step length have developed mathematical models that examine contrac-
combination, which they attributed to elastic energy storage tion energetics at the level of individual cross bridges (for
and return in the Achilles tendon. These results highlight review, see (24)). The advantage of this approach is that the
that net joint workYbased measures of mechanical work parameters in the model have direct links to the underlying
likely underestimate positive work because of factors such as biological processes. However, many of the values for these
inevitable cocontraction of antagonist muscles. It should be parameters are not well known, especially for humans. Un-
noted that our estimates for efficiency are high relative to the fortunately, these models do not lend themselves to inclusion
traditional values of 0.20Y0.30 measured in isolated muscles in simulation analyses because of their complexity, and they
or muscle fibers (e.g., (24)). This discrepancy could be caused are not easily integrated with the relatively simple and com-
by a number of factors associated with the model such as putationally efficient Hill-type muscle models commonly
unaccounted or underestimated elasticity in the muscle used in forward dynamics simulations. Because of this limi-
fibers, tendon, and other structures; an overestimate of the tation, a number of phenomenological models (e.g., (2,28))
energy lost at foot-ground contact; an overestimate of the have been developed that calculate metabolic cost based on
resting baseline metabolic cost used to determine net meta- the outputs from Hill-type muscle models.
bolic cost; unmodeled stretch-induced force enhancement; or To estimate the total energy used by a muscle during con-
the fact that the efficiency measured in isolated muscle fibers tractions, these phenomenological models include not only
is not the same as whole-body efficiency. This later point the mechanical power, but also the rate that heat is produced
has been the topic of considerable debate, and future work (Fig. 4). Parameters common to such models include
is needed to better understand efficiency in the context of activation heat rate, maintenance heat rate, shortening heat
whole-body movements (see target article by van Ingen rate, and mechanical work rate (e.g., (2,8,28)), whereas

208 Exercise and Sport Sciences Reviews [Link]

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
some also include basal metabolic rate (2) or heat dissipated to the experimental observations, 3) energy/momentum bal-
in elastic structures (8). Obtaining appropriate values for ances are assured, and 4) the results are insensitive to mo-
these parameters remains one of the primary challenges for derate variations in model parameters. Note that to achieve
use in simulation analyses of human movement. Few of these appropriate timing of muscle excitation, the timing in the
parameters can be measured directly in humans and are often optimization can be constrained to closely match measured
estimated from measurements made on isolated muscles (or EMG timing (e.g., Fig. 1). Comparisons with experimental
muscle fibers) from animals such as mice or frogs. Yet, of observations (e.g., joint motion, ground reaction forces, and
the studies that have compared simulated results with exist- inverse dynamicsYbased quantities) can be used to confirm
ing empirical metabolic cost data (2,27,28), there has been that the overall mechanics of the movement are sufficiently
relatively good agreement at multiple levels of complexity similar to the experimental observations (Fig. 1), and this is
(e.g., isolated muscles, single joint or whole-body move- inherent in the process of the optimal tracking we typically
ments). However, measurements using techniques such as use. Further comparisons can be made using experimental
near infrared spectroscopy or magnetic resonance spectros- techniques such as magnetic resonance imaging to confirm
copy will be essential to further refine and validate the models. and refine the musculoskeletal model geometry, muscle ar-
Because humans and other animals generally choose to chitecture and mechanics, and joint kinematics (for review,
perform tasks using mechanics that reduce metabolic en- see (3)), and high-speed ultrasound to verify muscle-tendon
ergy consumption, muscle energetics models have frequently interactions and elastic energy storage and return (e.g.,
been used in optimization objective functions in simulation (7,11)). Such techniques will be essential in future studies to
studies. Applications have ranged from using metabolic cost validate the model across multiple scales of the musculoske-
models to examine motor unit recruitment patterns (8) and letal system to aid in model development and validation and
the influence of muscle fiber-type composition on metabolic ultimately improve the accuracy and use of modeling and
cost in pedaling (29) to evaluating alternative gait strategies simulation techniques.
(23) and predicting overall walking mechanics (1). Studies However, achieving a close comparison between all exper-
such as these provide insight into the relationship between imental and simulation data can be difficult in part because
metabolic cost and an individual’s morphology or strategies experimental data have their own limitations, which are
used to perform a given locomotor task. often not acknowledged or tested for explicitly. For example,
As the estimates of individual parameters become more there is always some degree of error (both marker registration
accurate, so too will the models and simulations of muscle errors and skin movement artifact errors) associated with the
energetics. Current models remain an excellent tool for ex- experimental body segment kinematics that a kinematically
amining relative differences between muscles (e.g., pro- consistent simulation model will have difficulty tracking
portions of fiber types) or experimental conditions (e.g., (e.g., (4)). Joint momentYbased quantities derived from in-
differences in speed or load). Future studies may use muscle verse dynamics often have high residuals at the terminal link
energetics models to tease out the timing and distribution of because of dynamic inconsistency between the GRF and
energy consumption among muscles or to examine the effi- body segment kinematics (e.g., (20)). This results in exper-
ciency of individual muscles during various movement tasks. imental mechanical work quantities that are not dynamically
In addition, such models can be combined with forward dy- consistent, with the errors being difficult to identify because
namics simulations and design optimization to help improve there is not an independent method available to validate
rehabilitation outcomes. For example, orthotic or prosthetic joint work quantities because the amount of energy dissi-
devices can be optimized to minimize metabolic cost during pated in the musculoskeletal system (e.g., viscous damping in
walking or running. Thus, the development of accurate and body tissues) and in the environment (e.g., foot-ground
computationally efficient models has a wide range of pro- contact) is not known. In addition, the simulation data are
mising applications that make this a fruitful area for con- generally compared with group average data, which are often
tinued research. heavily smoothed and represent theoretical trajectories that
an individual human subject may not be able to produce in a
single walking trial. The advantage of forward dynamics
simulations is that they are dynamically consistent and satisfy
POTENTIAL LIMITATIONS OF SIMULATION the fundamental laws of physics at each point in the gait
ANALYSES cycle while being constrained to be consistent with our cur-
rent models of the muscle force-length-velocity-activation
Simulation studies face many challenges, with establishing relationships. Despite these challenges, a number of carefully
validity being one of the primary challenges because many of designed experimental studies have recently confirmed dif-
the quantities being investigated cannot be measured V ferent aspects of the simulation results (9,15,25). For ex-
hence, the reason for using simulation in the first place. ample, McGowan et al. (15) designed an experimental study
Modeling requires assumptions regarding anatomy, muscle that manipulated body weight and body mass independently
physiology, and structural and mechanical properties includ- to confirm the predictions from computer simulation studies
ing the interaction between the model and the ground. that the uniarticular and biarticular ankle plantar flexors
However, if quantities such as muscle work are of interest, have distinct functional roles in walking (i.e., although both
confidence in the results can be gained when 1) each muscle the gastrocnemius and soleus contribute to body support, the
is excited at the appropriate point in the gait cycle, 2) the soleus is the primary contributor to trunk forward propul-
overall mechanics of the movement are sufficiently similar sion). Similarly, Hernández et al. (9) used muscle electrical

Volume 37 c
Number 4 c
October 2009 Muscle Work in Human Locomotion 209

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.
stimulation to confirm the nonintuitive simulation predic- tendon interaction and elastic energy usage in human walking. J. Appl.
tion that the rectus femoris acts to accelerate the hip into Physiol. 2005; 99(2):603Y8.
12. Kautz SA, Hull ML, Neptune RR. A comparison of muscular
extension rather than flexion during the early swing phase of mechanical energy expenditure and internal work in cycling. J. Biomech.
walking. Such studies are critically needed to further validate 1994; 27(12):1459Y67.
the use of simulation as a powerful tool to analyze various 13. Kautz SA, Neptune RR. Biomechanical determinants of pedaling
aspects of human locomotion. energetics: Internal and external work are not independent. Exerc. Sport
Sci. Rev. 2002; 30(4):159Y65.
14. Kuo AD, Donelan JM, Ruina A. Energetic consequences of walking like
an inverted pendulum: Step-to-step transitions. Exerc. Sport Sci. Rev.
SUMMARY 2005; 33(2):88Y97.
15. McGowan CP, Neptune RR, Kram R. Independent effects of weight and
Muscle mechanical work is an important biomechanical mass on plantar flexor activity during walking: Implications for their
contributions to body support and forward propulsion. J. Appl. Physiol.
quantity in analyses of human locomotion; however, current 2008; 105(2):486Y94.
experimental methods are limited in their ability to quantify 16. Neptune RR, Sasaki K, Kautz SA. The effect of walking speed on muscle
it accurately. Muscle-actuated forward dynamics simulations function and mechanical energetics. Gait Posture. 2008; 28(1):135Y43.
provide a consistent mechanical solution that can be in- 17. Neptune RR, van den Bogert AJ. Standard mechanical energy analyses
terrogated at multiple levels (muscle fiber, musculotendon, do not correlate with muscle work in cycling. J. Biomech. 1998; 31(3):
239Y45.
net joint moment, and whole-body work) to gain insight into 18. Neptune RR, Zajac FE, Kautz SA. Muscle mechanical work require-
mechanical work requirements of human locomotion and ments during normal walking: The energetic cost of raising the body’s
better understand the limitations associated with experimen- center-of-mass is significant. J. Biomech. 2004; 37(6):817Y25.
tally derived measures of muscle work. Future work integrat- 19. Prilutsky BI. Work, energy expenditure, and efficiency of the stretch-
shortening cycle. J. Appl. Biomech. 1997; 13(4):466Y70.
ing detailed and computationally efficient metabolic cost
20. Remy CD, Thelen DG. Optimal estimation of dynamically consistent
models with forward dynamics simulations holds great kinematics and kinetics for forward dynamic simulation of gait.
promise to investigate individual muscle and whole-body J. Biomech. Eng. 2009; 131(3):031005.
energetics and efficiency across a wide range of locomotor 21. Sasaki K, Neptune RR, Kautz SA. The relationships between muscle,
tasks and patient populations. external, internal and joint mechanical work during normal walking.
J. Exp. Biol. 2009; 212(Pt 5):738Y44.
22. Sawicki GS, Ferris DP. Powered ankle exoskeletons reveal the metabolic
Acknowledgments cost of plantar flexor mechanical work during walking with longer steps
at constant step frequency. J. Exp. Biol. 2009; 212(Pt 1):21Y31.
23. Sellers WI, Dennis LA, W-J W, Crompton RH. Evaluating alternative
The authors thank Dr. Kotaro Sasaki for his contributions to the more gait strategies using evolutionary robotics. J. Anat. 2004; 204(5):343Y51.
recent simulation work. This work was supported by NIH Grant R01 24. Smith NP, Barclay CJ, Loiselle DS. The efficiency of muscle contrac-
NS55380 and the Rehabilitation Research and Development Service of the tion. Prog. Biophys. Mol. Biol. 2005; 88(1):1Y58.
Department of Veterans Affairs. 25. Stewart C, Postans N, Schwartz MH, Rozumalski A, Roberts A. An
exploration of the function of the triceps surae during normal gait using
functional electrical stimulation. Gait Posture 2007; 26(4):482Y8.
References 26. Thelen DG, Chumanov ES, Sherry MA, Heiderscheit BC. Neuro-
musculoskeletal models provide insights into the mechanisms and
1. Anderson FC, Pandy MG. Dynamic optimization of human walking. rehabilitation of hamstring strains. Exerc. Sport Sci. Rev. 2006; 34(3):
J. Biomech. Eng. 2001; 123(5):381Y90. 135Y41.
2. Bhargava LJ, Pandy MG, Anderson FC. A phenomenological model for 27. Umberger BR. The Cost of Swinging the Leg in Human Walking. American
estimating metabolic energy consumption in muscle contraction. Society of Biomechanics Annual Conference. Blacksburg (VA): American
J. Biomech. 2004; 37(1):81Y8. Society of Biomechanics Annual Conference Proceedings CD; 2006.
3. Blemker SS, Asakawa DS, Gold GE, Delp SL. Image-based muscu- 28. Umberger BR, Gerritsen KG, Martin PE. A model of human muscle
loskeletal modeling: Applications, advances, and future opportunities. energy expenditure. Comput. Methods Biomech. Biomed. Engin. 2003;
J. Magn. Reson. Imaging 2007; 25(2):441Y51. 6(2):99Y111.
4. Cahouët V, Luc M, David A. Static optimal estimation of joint accele- 29. Umberger BR, Gerritsen KG, Martin PE. Muscle fiber type effects on
rations for inverse dynamics problem solution. J. Biomech. 2002; 35(11): energetically optimal cadences in cycling. J. Biomech. 2006; 39(8):
1507Y13. 1472Y9.
5. Doke J, Donelan JM, Kuo AD. Mechanics and energetics of swinging 30. Umberger BR, Martin PE. Mechanical power and efficiency of level
the human leg. J. Exp. Biol. 2005; 208(Pt 3):439Y45. walking with different stride rates. J. Exp. Biol. 2007; 210(Pt 18):
6. Donelan JM, Kram R, Kuo AD. Mechanical work for step-to-step 3255Y65.
transitions is a major determinant of the metabolic cost of human 31. van Ingen Schenau GJ, Bobbert MF, de Haan A. Does elastic energy
walking. J. Exp. Biol. 2002; 205(Pt 23):3717Y27. enhance work and efficiency in the stretch-shortening cycle? J. Appl.
7. Fukunaga T, Kubo K, Kawakami Y, Fukashiro S, Kanehisa H, Maganaris Biomech. 1997; 13(4):389Y415.
CN. In vivo behaviour of human muscle tendon during walking. Proc. 32. van Ingen Schenau GJ, van Woensel WW, Boots PJ, Snackers RW,
Biol. Sci. 2001; 268(1464):229Y33. de Groot G. Determination and interpretation of mechanical power
8. Hatze H, Buys JD. Energy-optimal controls in the mammalian neuro- in human movement: Application to ergometer cycling. Eur. J. Appl.
muscular system. Biol. Cybern. 1977; 27(1):9Y20. Physiol. Occup. Physiol. 1990; 61(1Y2):11Y9.
9. Hernandez A, Dhaher Y, Thelen DG. In vivo measurement of dynamic 33. Willems PA, Cavagna GA, Heglund NC. External, internal and total
rectus femoris function at postures representative of early swing phase. work in human locomotion. J. Exp. Biol. 1995; 198:(Pt 2):379Y93.
J. Biomech. 2008; 41(1):137Y44. 34. Winter DA. A new definition of mechanical work done in human
10. Houdijk H, Pollmann E, Groenewold M, Wiggerts H, Polomski W. The movement. J. Appl. Physiol. 1979; 46(1):79Y83.
energy cost for the step-to-step transition in amputee walking. Gait 35. Zajac FE, Neptune RR, Kautz SA. Biomechanics and muscle coordina-
Posture. 2009; 30(1):35Y40. tion of human walking. Part I: Introduction to concepts, power transfer,
11. Ishikawa M, Komi PV, Grey MJ, Lepola V, Bruggemann GP. Muscle- dynamics and simulations. Gait Posture. 2002; 16(3):215Y32.

210 Exercise and Sport Sciences Reviews [Link]

Copyright @ 2009 by the American College of Sports Medicine. Unauthorized reproduction of this article is prohibited.