Biomechanics and energetics of walking on uneven terrain

Abstract

Walking on uneven terrain is more energetically costly than walking on smooth ground, but the biomechanical factors that contribute to this increase are unknown. To identify possible factors, we constructed an uneven terrain treadmill that allowed us to record biomechanical, electromyographic, and metabolic energetics data from human subjects. We hypothesized that walking on uneven terrain would increase step width and length variability, joint mechanical work, and muscle co-activation compared to walking on smooth terrain. We tested healthy subjects (N=11) walking at 1.0 m/s, and found that, when walking on uneven terrain with up to 2.5 cm variation, subjects decreased their step length by 4% and did not significantly change their step width, while both step length and width variability increased significantly (22% and 36%, respectively; p<0.05). Uneven terrain walking caused a 28% and 62% increase in positive knee and hip work, and a 26% greater magnitude of negative knee work (0.0106, 0.1078, and 0.0425 J/kg, respectively; p<0.05). Mean muscle activity increased in seven muscles in the lower leg and thigh (p<0.05). These changes caused overall net metabolic energy expenditure to increase by 0.73 W/kg (28%; p<0.0001). Much of that increase could be explained by the increased mechanical work observed at the knee and hip. Greater muscle co-activation could also contribute to increased energetic cost but to unknown degree. The findings provide insight into how lower limb muscles are used differently for natural terrain compared to laboratory conditions.

Full text

3963
INTRODUCTION
Humans and other animals navigate complex terrain in their
everyday lives. From uneven sidewalks to natural trails, humans
often encounter surfaces that are not smooth. Energetic cost for
locomotion increases on natural complex surfaces [e.g. grass, sand,
snow (e.g. Davies and Mackinnon, 2006; Pandolf et al., 1976;
Pinnington and Dawson, 2001; Soule and Goldman, 1972)]
compared with smooth surfaces, but the biomechanical mechanisms
responsible for the increased cost are still unclear. Terrain has many
features that might affect locomotion, such as height variations,
damping and coefficient of friction. These could cause a variety of
changes to locomotion, yet gait research has typically focused on
smooth, level ground. To provide some insight into how complex
natural terrain can affect locomotion, we studied metabolic energy
expenditure and biomechanics of human walking on a synthesized
uneven terrain surface.
There are a number of potential factors that could contribute to
greater energy expenditure when walking on uneven terrain
compared with smooth terrain. Adjusting step parameters during
locomotion is one such factor. Adults typically take shorter and wider
steps with increasing age (Murray et al., 1969), while younger
individuals respond similarly to continuous perturbations, both
physical and visual (Hak et al., 2012; McAndrew et al., 2010). If
these are strategies to enhance stability, it is possible that younger
adults might do the same on uneven terrain. Such terrain may also
perturb gait from step to step and cause greater variability. Step
width, in particular, could show increased variability, because lateral
balance may be more dependent on active stabilization than fore–aft
motion, because of passive dynamic stability (Donelan et al., 2001).
Energy expenditure would be expected to increase with changes in
mean step parameters (Gordon et al., 2009; Wade et al., 2010) and
with changes in step variability as well (O’Connor et al., 2012).
Uneven terrain might also require more mechanical work from
the legs, independent of the effect on step parameters. Kuo (Kuo,
2002) previously hypothesized that walking economy is improved
by pushing off with the trailing leg just prior to the collision of the
leading leg. Push-off redirects the body center of mass and, if
properly timed, can reduce the amount of negative work performed
in the collision. Uneven terrain may upset the relative timing of
these events, so that a collision occurring either earlier or later
relative to push-off would be expected to lead to greater negative
mechanical work. This would then require muscles to compensate
and actively do more positive work elsewhere, as steady walking
requires zero work on average. It is difficult to predict how work
will be distributed between the lower limb joints, but perturbed
timing would be expected to require more work overall, and thus
more expenditure of metabolic energy.
Another possible factor that could contribute to increased energy
expenditure is co-activation of muscles. When walking on less secure
surfaces such as railroad ballast or ice (Cappellini et al., 2010;
Marigold and Patla, 2002; Wade et al., 2010), or when there is an
unexpected drop in the surface (Nakazawa et al., 2004), humans
increase muscle co-activation about the ankle joint. This compensation
may help to stabilize the joints for uncertain conditions. If humans
co-activate the corresponding muscles on uneven terrain, energy
expenditure may increase even if work does not.
SUMMARY
Walking on uneven terrain is more energetically costly than walking on smooth ground, but the biomechanical factors that
contribute to this increase are unknown. To identify possible factors, we constructed an uneven terrain treadmill that allowed us
to record biomechanical, electromyographic and metabolic energetics data from human subjects. We hypothesized that walking
on uneven terrain would increase step width and length variability, joint mechanical work and muscle co-activation compared with
walking on smooth terrain. We tested healthy subjects (N=11) walking at 1.0
ms
–1
, and found that, when walking on uneven terrain
with up to 2.5
cm variation, subjects decreased their step length by 4% and did not significantly change their step width, while
both step length and width variability increased significantly (22 and 36%, respectively; P<0.05). Uneven terrain walking caused a
28 and 62% increase in positive knee and hip work, respectively, and a 26% greater magnitude of negative knee work (0.0106,
0.1078 and 0.0425
Jkg
–1
, respectively; P<0.05). Mean muscle activity increased in seven muscles in the lower leg and thigh
(P<0.05). These changes caused overall net metabolic energy expenditure to increase by 0.73
Wkg
–1
(28%; P<0.0001). Much of that
increase could be explained by the increased mechanical work observed at the knee and hip. Greater muscle co-activation could
also contribute to increased energetic cost but to unknown degree. The findings provide insight into how lower limb muscles are
used differently for natural terrain compared with laboratory conditions.
Key words: energetics, joint work, kinematics, uneven terrain.
Received 3 December 2012; Accepted 15 July 2013
The Journal of Experimental Biology 216, 3963-3970
©2013. Published by The Company of Biologists Ltd
doi:10.1242/jeb.081711
RESEARCH ARTICLE
Biomechanics and energetics of walking on uneven terrain
Alexandra S. Voloshina
1,2,
*, Arthur D. Kuo
2
, Monica A. Daley
3
and Daniel P. Ferris
1
1
School of Kinesiology, University of Michigan, Ann Arbor, MI, USA,
2
Department of Mechanical Engineering, University of Michigan,
Ann Arbor, MI, USA and
3
Comparative Biomedical Sciences, The Royal Veterinary College, London, UK
*Author for correspondence (voloshis@umich.edu)
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3964
The purpose of this study was to determine the changes in walking
biomechanics on uneven terrain, and how they might relate to
increased metabolic cost. We developed an uneven terrain surface
that allowed us to collect continuous kinematic and energetics data
during treadmill and over-ground walking. We expected that
walking on uneven terrain would increase the variability of step
width and step length. Humans may also adopt wider and shorter
steps as a stabilizing strategy, similar to the changes that older adults
make to compensate for poorer balance. Regardless of strategy, the
perturbations of uneven terrain would be expected to cause subjects
to increase joint mechanical work and muscle co-activation on
uneven terrain compared with walking on smooth terrain. Walking
over natural surfaces involves much greater variation than a smooth
treadmill belt or uniform pavement; thus, biomechanics and
energetics in uneven terrain are likely to better represent the
functional demands that have influenced the evolution of human
bipedalism (Pontzer et al., 2009; Sockol et al., 2007).
MATERIALS AND METHODS
We created an uneven terrain surface by attaching wooden blocks
to a treadmill belt. This allowed us to collect biomechanical data
and metabolic energetics data simultaneously during continuous
walking. The same terrain surface could also be placed over ground-
embedded force plates, facilitating collection of joint kinetics data.
Each wooden block was covered with a layer of ethylene-vinyl
acetate (EVA) cushioning foam, to make the surface comfortable
to walk on. To test for effects of the cushioning foam alone, subjects
also walked on a smooth treadmill belt surface covered only by the
cushioning foam, resulting in conditions termed ‘uneven + foam’
and ‘even + foam’. We also tested walking on just the normal
treadmill belt, termed the ‘even’ condition. We collected kinematic,
kinetic, metabolic and electromyographic data for each condition,
all at a walking speed of 1.0ms
–1
.
Subjects
Eleven young, healthy subjects (four female, seven male, mean ±
s.d.: age 22.9±2.8years, mass 66.1±13.2kg and height 172.6±6.4cm)
participated in the study. Data were collected in two sessions on
separate days. One session was for treadmill walking to collect
oxygen consumption (N=7), step parameter data (N=9) and
electromyographic data (N=8). The other session was for over-
ground walking over force plates to collect joint kinematics and
kinetics (N=10) data. Some data were not collected successfully
because of technical and logistical issues, resulting in values of N
less than 11 in each data subset, noted in parentheses above. Because
of these issues, different subject data were excluded from step
parameter, kinematic, kinetic and electromyographic data.
Subjects provided written informed consent before the
experiment. All procedures were approved by the University of
Michigan Health Sciences Institutional Review Board.
Walking surfaces and trial procedures
We modified a regular exercise treadmill (JAS Fitness Systems,
Trackmaster TMX22, Dallas, TX) to allow for attachment and
replacement of uneven and even terrain surfaces (Fig.1). The uneven
surface was created from wooden blocks arranged in squares
(15.2×15.2cm) and glued together to form three different heights
(1.27, 2.54 and 3.81cm) and create an uneven surface (after
Sponberg and Full, 2008). Each square consisted of smaller blocks,
2.55×15.2cm, oriented lengthwise across the belt and affixed to it
with hook-and-loop fabric. The short dimension of the blocks
allowed the belt to curve around the treadmill rollers. Each block’s
surface was covered with a layer of cushioning foam that was
1.27cm thick, yielding the uneven + foam surface condition. Even
though the uneven squares were arranged in a repeating pattern,
their length was not an integer fraction of step length, making it
difficult for subjects to learn or adopt a periodic compensation for
this condition.
The two other surfaces served as control conditions. The even +
foam condition was formed using only cushioning foam of the same
height as the uneven + foam condition. The even condition consisted
of the treadmill belt alone, and allowed us to determine the
biomechanical effects of only the cushioning foam.
Walking trials were performed for all three conditions in
randomized order, both on treadmill and over-ground. All trials were
completed with subjects walking at 1.0ms
–1
while wearing rubber-
soled socks for comfort. Subjects were instructed to walk naturally
and encouraged not to look down at their feet unless they felt
unstable. Subjects participated in only one 10min treadmill trial per
condition with at least 5min of resting time between trials. During
over-ground trials, speed was verified by optical timers set 4m apart
mid-way in a 7m path, and trials were only used if they were within
10% of the target time. Subjects completed at least 10 successful
over-ground trials for each surface condition.
Kinetics and kinematics
For all walking trials (both on the treadmill and over-ground), we
recorded the position of 31 reflective markers located on the pelvis
and lower limbs using a 10-camera motion capture setup (frame
rate: 100Hz; Vicon, Oxford, UK). Markers were taped to the skin
or spandex shorts worn by the subjects. Three markers were placed
on each thigh and shank, one at the sacrum and one at each of the
greater trochanters, anterior superior iliac spine, the medial and
lateral epicondyles of the femur, the medial and lateral malleoli, the
fifth metatarsals, the calcanei, and the first metatarsals. Medial
markers were removed after static marker calibration. Only the last
2.5min of kinematic data collected from each treadmill trial were
used for calculations. Over-ground trials occurred over two force
plates, yielding one to two steps per trial for inverse dynamics
calculations. The marker data for both legs were low-pass filtered
at 6Hz to reduce motion artifact (fourth-order Butterworth filter,
The Journal of Experimental Biology 216 (21)
W
L
H
Foam
Wood
A
B
C
B
C
Fig.1. (A)Treadmill with the uneven
terrain surface attached. (B)
Schematic
of the uneven surface layout,
consisting of three alternating heights
(arrows indicate the treadmill’s long
axis). (C)
Close-up representation of
the individual blocks comprising each
stepping area. Dimensions: H, 1.27
cm;
L, 15.2
cm; W, 2.54cm.
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3965Mechanics of walking on uneven terrain
zero-lag), and used to calculate step widths, lengths and heights, as
well as to identify successful steps in over-ground trials. Step
parameters were calculated using the calcaneous markers on the
two feet. Step width and length were defined as the lateral and
fore–aft distances between the two markers at their respective heel-
strike instants. Step height was defined as the vertical distance
between the two markers at heel-strike, and was only used to indicate
greater step height variability expected from uneven terrain. Heel-
strike was defined by the onset of ground force for over-ground
trials, and by the lowest height of the calcaneous marker for treadmill
trials (where forces were not measured). Over-ground data were
used to confirm that these timings agreed well with each other. All
step measurements were normalized to subject leg length, defined
as the average vertical distance between the greater trochanter and
calcaneous markers of both legs.
The uneven + foam and even + foam surfaces could be detached
from the treadmill and used as a walkway. During over-ground trials,
subjects walked across these two walking surfaces placed on top of
two in-ground force platforms, 0.5m apart (sample rate: 1000Hz;
AMTI, Watertown, MA, USA) for the uneven + foam and even +
foam conditions. The surfaces were not secured to the floor, but did
not appear to slip during walking trials. For the even condition,
subjects walked on the bare floor and force plates. The in-ground
force plates were re-zeroed between conditions. All force data were
low-pass filtered at 6Hz (fourth-order Butterworth filter, zero lag)
and ground reaction force data were synchronized with the kinematic
data. Joint angles, moments and powers for the stance limb were
determined using inverse dynamics analysis in Visual-3D (C-
Motion, Germantown, MD, USA). Positive and negative joint work
measures were calculated by integrating the intervals of either
positive or negative joint power over time.
Electromyography
We measured electromyography (EMG) in the tibialis anterior (TA),
soleus (SO), medial gastrocnemius (MG), lateral gastrocnemius
(LG), rectus femoris (RF), vastus medialis (VM), vastus lateralis
(VL) and the semitendinosus of the medial hamstring (MH) muscles,
during all treadmill trials. All EMG data were collected only for
the right leg. Bipolar surface electrodes (sample rate: 1000Hz;
Biometrics, Ladysmith, VA, USA) were placed over the belly center
of the muscle and in parallel to the muscle according to the procedure
of Winter and Yack (Winter and Yack, 1987). The inter-electrode
distance was 2.0cm for all trials and electrode diameters were 1.0cm.
The EMG amplifier had a bandwidth of 20–460Hz. As with other
measurements, only the last 2.5min of EMG data were used for
data analysis. All EMG signals were high-pass filtered with a 20Hz
cut-off frequency (fourth-order Butterworth filter, zero-lag) and then
full-wave rectified. We then normalized each muscle’s data to the
maximum activation observed for that same muscle over all three
conditions for that subject (Winter and Yack, 1987; Yang and
Winter, 1984) and averaged over subjects to create representative
EMG profiles. Standard deviations of the EMG traces were found
at each time point for every subject and condition and also averaged,
to determine mean standard deviation envelopes. Although the
relationship between EMG variability and metabolic cost is
undetermined, this measure can indicate the level of perturbation
to gait mechanics from uneven terrain. To determine increases in
muscle activation, we found the average of the normalized EMG
profile for each subject and condition. These average values were
then averaged over subjects. In addition, we assessed muscle co-
activation as the amount of mutual contraction (MC) as defined by
Thoroughman and Shadmehr (Thoroughman and Shadmehr, 1999)
to indicate ‘wasted’ contraction, for each stride for three pairs of
antagonistic muscles (SO/TA, MH/VM and MH/VL). To do so, we
used the equation:
where f
1
and f
2
are the full-wave rectified EMG profiles, averaged
over 100 steps, of the two antagonistic muscles, and min(f
1
, f
2
) is
the minimum of the two profiles at each time point. Integrals were
computed over the duration of the whole stride and in 1% increments
to identify where in the stride cycle mutual contraction occurred.
Metabolic rate
For all treadmill walking conditions, we measured the rate of oxygen
consumption (V
O
2
) using an open-circuit respirometry system
(CareFusion Oxycon Mobile, Hoechberg, Germany). We recorded
7min of respirometry data during a quiet standing trial, and 10min
for all walking trials. Although 3-min trials are sufficient to reach
steady-state energy expenditure on uniform terrain (Poole and
Richardson, 1997), we expected walking on uneven terrain to be
an increase in exercise intensity and allowed subjects 7.5min of
walking to reach steady-state before collection 2.5min of data. We
later confirmed that subjects had reached steady-state in both
biomechanics and energetics on the novel terrain conditions by
checking that no adaptation trends were still present in the last
2.5min of data. We calculated the metabolic energy expenditure
rate of each subject using standard empirical equations yielding
metabolic rate E
met
(W) (Brockway, 1987; Weir, 1949). Net
metabolic rate was calculated by subtracting the standing metabolic
power from the metabolic power of all other conditions. We
normalized the net metabolic power for all conditions by dividing
by subject body mass (kg).
Data and statistical analyses
To compare changes in variability for step parameter, joint parameter
and EMG data, we averaged the variability for each of the three
conditions over all subjects. For step data, we defined variability
as the standard deviation of contiguous step distances or periods
over time for each subject. For joint parameter and EMG data, means
were calculated across trials for each point in relative stride cycle
timing. Similarly, joint parameter and EMG variability was defined
for each subject and condition as the standard deviation across trials
for each point. We then reported the mean variations (and standard
deviations) across subjects for each condition. Differences between
the conditions were quantified by performing repeated-measures
ANOVAs on the data sets of interest. The significance level α was
set at 0.05 and post hoc Holm–Sidak multiple comparison tests were
performed where appropriate.
RESULTS
Walking on uneven terrain resulted in a variety of changes to gait
compared with walking on smooth terrain. Subjects walked with
slightly shorter step lengths and substantially increased step
variability. Gait kinematics remained similar overall, but knee and
hip mechanical work increased on uneven terrain. We also
observed increased mean activity among multiple proximal leg
muscles (VM, VL, RF, MH), and greater muscle mutual
contraction about all three joints on uneven terrain. In all variables,
the two smooth terrain conditions (with and without a foam layer)
exhibited negligible differences between each other. We therefore
report comparisons mainly between the uneven + foam and even
+ foam conditions.
ff tMC min( , )d , (1)
12
∫
=
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3966
Kinetics and kinematics
Although mean step parameters changed little, there were large
changes in step variability during walking on the uneven surface
when compared with the even foam surface (Table1). Of the mean
step distances, only step length changed significantly, decreasing
by 3.7%. Because walking speed was kept fixed, this was
accompanied by a 3.7% decrease in mean step duration. Variability
of step width, length and height all increased significantly by
approximately 35, 23 and 105%, respectively. Step period variability
also increased significantly by 26.7%.
A number of effects were observed on joint kinematics and
kinetics when subjects walked on uneven terrain when compared
with the even surface (Fig.2). Qualitative examination of sagittal
plane joint angles on uneven terrain suggest slightly greater knee
and hip flexion at mid-swing, perhaps associated with greater
ground clearance of the swing foot. Mean ankle angle trajectory
changed little (Fig.2). However, on uneven terrain, we observed
larger effects on the joint moments during stance, with increased
knee flexion and increased hip extension moments at mid-stance.
At the end of stance during push-off, these patterns reversed, with
greater knee extension and hip flexion moments. The main
changes in joint power were also confined to the knee and hip,
with increased peak powers, especially at push-off (by
approximately 65 and 85%, respectively) when walking on the
uneven surface. Hip power also increased by 75% during mid-
stance, at approximately 20% of stride time. Toe-off timing in
the stride cycle did not appear to differ between conditions. Joint
trajectories were more variable on uneven terrain (Fig.2). The
The Journal of Experimental Biology 216 (21)
Table1. Step parameters for three terrain conditions
Even Even + foam Uneven + foam P
Mean Step variability Mean Step variability Mean Step variability Mean Step variability
Width 0.077±0.040 0.027±0.005 0.080±0.036 0.028±0.004 0.102±0.053 0.038±0.006* 0.0336 0.0003
Length 0.672±0.020 0.037±0.009 0.662±0.025 0.037±0.008 0.638±0.024* 0.045±0.007* 0.0039 0.0006
Height – 0.004±0.001 – 0.004±0.001 – 0.008±0.001* – <0.0001
Step period (s) 0.568±0.022 0.013±0.003 0.560±0.027 0.014±0.003 0.540±0.038* 0.018±0.003* 0.0028 0.0017
Parameters include mean step length, width and height and their respective variations (all normalized to subject leg length, mean 0.870m), as well as step
period. Values are means ± s.d. across subjects. Step variability is defined as the standard deviation of step distances over a trial, reported as a mean
(±s.d.) across subjects. Asterisks signify a statistically significant difference of the uneven + foam condition from the other two conditions (post hoc pair-wise
comparisons, α=0.05).
0 50 100
–20
0
20
0 0
–50
0
100
150
0
–50
0
0
–20
0
20
0
–150
–100
–50
0
50
0
–40
–20
0
0
–50
0
50
0
0
50
100
50 100 50 100
50 100 50 100 50 100
50 10050 10050 100
Joint angle (deg)
Joint torque (N-m)
Joint power (W kg
–1
)
Ankle
Knee
Hip
% Stride time
Even + foam
Uneven + foam
Ext.
Flex.
Ext.
Flex.
0
100
50
–25
–75
50
Fig.2. Joint angle, torque and power versus stride time for two terrain conditions. Mean trajectories for ankle, knee and hip are plotted against percent stride
time for uneven and even terrain (both with foam) conditions. Shaded area denotes standard deviation across subjects for uneven + foam; dashed lines for
even + foam. Strides start and end at same-side heel-strike; dashed vertical gray lines indicate toe-off.
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3967Mechanics of walking on uneven terrain
ankle angle variability more than doubled on uneven terrain, while
the knee and hip variability increased by ~30% (all P<0.05). The
mean ankle and knee torque variability both increased by ~50%
(all P<0.05). All joint power variability also increased by 50%
or more in the uneven terrain condition (all P<0.05).
The biomechanical effects included greater joint work performed
over a stride (Fig.3). There was a 0.0106Jkg
–1
(28%) increase in
positive knee work and a 0.0425Jkg
–1
(26%) increase in negative
knee work (P=0.011 and P=0.0019, respectively). Positive hip work
also significantly increased by 0.1078Jkg
–1
(62%; P<0.0001). No
statistically significant changes were found in positive or negative
ankle work, or negative hip work.
Muscle activation
Subjects showed increased muscle activity, variability of activity
(Fig.4) and mutual contraction when walking on the uneven surface.
There were significant increases in activation for six of the eight
muscles measured (Fig.5). Averaged, normalized EMG values
increased for all of the thigh muscles: VM, VL, RF and MH
increased by 49, 60, 54 and 47%, respectively (P<0.05). In the lower
leg, SO muscle activity increased by 28%, while the MG muscle
activity increased by 17% (P<0.05). The remaining muscles, TA
and LG, did not exhibit significant changes in mean activity across
the stride, although TA appeared to have slightly decreased activity
in the first 10% of stride.
Variability of EMG increased significantly for nearly all muscles
on the uneven terrain (Fig.4). On average, walking on uneven terrain
resulted in a larger increase in variability (standard deviation of
muscle activity) in the thigh muscles (mean 60% increase) than in
the leg muscles (mean 30% increase). For the thigh muscles, RF
and VL variability increased by over 80% (P<0.05), and VM and
MH muscles showed increases of over 45% (P<0.05). The SO, MG
and LG muscles in the leg showed a minimum increase in standard
deviation of 27%, and as much as 40% for MG (P<0.05).
We also observed changes in co-contraction over the entire stride
for all three pairs of antagonistic muscles (Table2). However, upon
breaking the stride down into 1% increments, mutual activation for
the MH/VM and MH/VL muscle pairs appears to increase
substantially only around mid-stance. The MH/VL muscle pair also
shows a significant increase pre toe-off. The largest increase of
mutual contraction of the TA/SO muscles was seen shortly after
heel-strike (Fig.4).
Metabolic energy expenditure
Walking on the uneven terrain resulted in a significant increase in
energy expenditure compared with the other surfaces (Fig.6). Net
metabolic rate increased from 2.65±0.373Wkg
–1
(mean ± s.d.) to
3.38±0.289Wkg
–1
(P<0.0001), approximately 28%, from the even
foam to uneven terrain. There was no difference between the
*
*
Positive work (J kg
–1
)
Ankle Knee Hip
*
Negative work (J kg
–1
)
Even + foam
Uneven + foam
Even
0.3
0.2
0.1
0
0
–0.1
–0.2
1
2
0
50 100
TA
1
2
0
50 100
MH
1
2
0
50 100
VL
1
2
0
50 100
VM
1
2
0
50 100
RF
1
2
0
50 100
LG
1
2
0
50 100
MG
1
2
0
50 100
SO
% Stride time
MH/VL
TA/SO
MH/VM
[ ]
Even + foam
Uneven + foam
0% 170%85%
EMG activity
Fig.3. Joint work per stride for three terrain conditions. Values shown are
positive and negative work for ankle, knee and hip, with error bars denoting
standard deviations. Dashed lines indicate net work for that specific joint
and condition. Asterisks signify a statistically significant difference of the
uneven + foam condition from the other two conditions (α=0.05).
Fig.
4. Averaged electromyographic
(EMG) activity versus stride time for
even and uneven terrain conditions.
EMG data were normalized to the
maximum activation of each muscle
for each subject and plotted against
percent stride time for uneven and
even terrain (both with foam).
Strides start and end at same-side
heel-strikes; dashed vertical gray
lines indicate toe-off. Envelopes
indicate standard deviations for
uneven (shaded area) and even
terrain (dashed lines) conditions
(both with foam). Gray bars indicate
statistically significant increases in
mutual muscle contraction, with
darker colors indicating larger
percent increases, from even terrain
mutual muscle contraction to
uneven terrain mutual muscle
contraction. Brackets indicate time
of decreased muscle contraction.
TA, tibialis anterior; SO, soleus; MG,
medial gastrocnemius; LG, lateral
gastrocnemius; VM, vastus medialis;
VL, vastus lateralis; RF, rectus
femoris; MH, medial hamstring.
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3968
energetic cost of walking on the even surface (mean metabolic rate
of 2.53±0.282Wkg
–1
) and the even foam surface (P=0.330). Mean
standing metabolic rate was found to be 1.48±0.181Wkg
–1
.
DISCUSSION
On natural terrain, there are many surface properties that can dictate
the metabolic cost of locomotion. Surface compliance and damping
can affect locomotion energetics and dynamics (Ferris et al., 1998;
Ferris et al., 1999; Kerdok et al., 2002), as do surface inclines or
declines (Margaria, 1976; Minetti et al., 1993). However, few studies
have characterized the biomechanics and energetics of walking on
uneven surfaces. We examined the effects of uneven terrain
compared with smooth surfaces, and found a number of
biomechanical factors related to energetic cost. Locomotion on
terrain with a surface variability of only 2.5cm resulted in a 28%
increase in net metabolic cost. For comparison, this is approximately
energetically equivalent to walking up a 2% steady incline
(Margaria, 1968) and is likely comparable to natural terrain variation
experienced when moving over trails, grass or uneven pavement.
We observed only modest changes in stepping strategy with
uneven terrain. For example, average step length decreased by only
4%, and the increase in step width was not significant. Examination
of previous studies on the effects of varying step parameters
(Donelan et al., 2001; Gordon et al., 2009; O’Connor and Kuo, 2009)
suggests that differences seen here are too small to have a substantial
influence on energetic cost. However, we did observe a 22% increase
in step length variability and a 36% increase in step width variability.
As shown by others (Donelan et al., 2004; O’Connor et al., 2012),
it is costlier to walk with more variability (e.g. 65% greater step
width variability results in 5.9% higher energetic cost), in part
because increased step variability reduces the use of passive energy
exchange and increases step-to-step transition costs. However, the
differences we found in our study are not likely to translate to large
changes in energetic cost. Available evidence suggests that changes
in step distances and variability could account for only a small
percent of increased energy expenditure.
One of the biomechanical effects that might explain the energetic
cost differences was the amount and distribution of work by lower
limb joints. Work performed by the ankle over a stride did not change
appreciably on the uneven surface, but the hip performed 62% more
positive work and the knee 26% more negative work (Fig.3). The
greater positive work at the hip occurred during mid-stance and also
at push-off, as corroborated by increased medial hamstring and
rectus femoris activity (Figs4, 5). The hip accounted for nearly all
of the increase in positive joint work. Changes in positive joint work
relative to changes in metabolic energy cost yield a delta efficiency
(ΔEff=ΔW
+
/ΔE, where W
+
is positive mechanical power and E is
metabolic power) of approximately 32% (Fig.7). If all of the
increased metabolic energy cost of walking on uneven terrain came
exclusively from positive muscle work, then ΔEff would equal ~25%
(Margaria, 1968). A very low efficiency would imply that energy
is expended for costs other than work, such as increased co-activation
and force of contraction. But the relatively high ΔEff observed here
The Journal of Experimental Biology 216 (21)
Table2. Muscle mutual contraction for the entire stride for three terrain conditions
Even Even + foam Uneven + foam P
TA/SO 115.5±25.59 121.6±28.48 161.3±38.70* 0.0003
MH/VM 97.82±40.31 103.3±44.82 145.5±52.82* 0.0061
MH/VL 102.8±26.08 107.4±33.69 165.6±40.41* 0.0002
Values signify unitless area under the minimum of the normalized EMG activation curves for the two muscles of interest. Three muscle antagonist pairs are
compared: tibialis anterior/soleus (TA/SO), medial hamstring/vastus medialis (MH/VM) and medial hamstring/vastus lateralis (MH/VL). Asterisks signify a
statistically significant difference of the uneven + foam condition from the other two conditions (post hoc pair-wise comparisons, α=0.05). Values are means
± s.d. Standard deviations indicate variation between subjects.
0.2
0.4
0
0.1
0.2
0.3
0
TA
MHRFVLVM
LGMGSO
*
Mean EMG (% max)
Even + foam
Uneven + foam
Even
*
*
*
*
*
*
Even Even +
foam
Uneven +
foam
0
1
2
3
4
Net metabolic rate (W kg
–1
)
Fig.5. Averaged rectified EMG values normalized to maximum muscle
activation. Bars indicate standard deviation across subjects. Single
asterisks denote statistically significant differences between the uneven +
foam condition and the other two conditions. No statistically significant
differences were found between the even and even + foam conditions
(α=0.05).
Fig.
6. Net metabolic rate for three terrain conditions. Metabolic rates are
normalized by subject mass. Values shown are means over subjects, with
error bars indicating standard deviations. Asterisk indicates a statistically
significant difference between the uneven + foam walking condition and the
other two conditions (α=0.05).
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3969Mechanics of walking on uneven terrain
suggests that the cost of walking on uneven terrain may largely be
explained by greater mechanical work, mostly performed at the hip.
By exceeding 25% ΔEff, the data also suggest that not all of the
changes in joint positive work were due to active muscle work. Joint
power trajectories (Fig.2) reveal that some of the positive hip work
was performed simultaneously with negative knee work at toe-off
(at ~60% of stride time). The rectus femoris muscle is biarticular
and can flex the hip and extend the knee at the same time. It can
thus produce both higher positive work at one joint and a greater
negative work at the other, yet experience a smaller change in actual
muscle work. In addition, some joint work may be performed
passively through elastic energy storage and return by tendon, as
has been implicated most strongly for the ankle (Sawicki et al., 2009)
but also in the knee and hip (Doke and Kuo, 2007; Geyer et al.,
2006). It is therefore likely that positive joint work is an overestimate
of actual muscle work, which could explain the relatively high ΔEff.
It is nevertheless evident that there was substantially more positive
work at the hip, even discounting hip power at toe-off. The work
increase in the first half of stride is not easily explained by
simultaneous negative work at another joint, nor by passive elastic
work. It therefore appears that much of the increase in metabolic
cost could still be explained by active joint work, at a more
physiological efficiency.
A possible explanation for the joint work increase on uneven
terrain is the timing of push-off and collision during walking. Push-
off by the trailing leg can reduce negative work done by the leading
leg if it commences just before heel-strike, redirecting the body
center of mass prior to collision (Kuo, 2002; Kuo et al., 2005). Stride
period was quite consistent on level ground, with variability of
~0.014s, but increased by ~27% on uneven terrain. This may suggest
greater variability in timing between push-off and collision, which
may contribute to greater variability of joint power and muscle
activity to compensate for collision costs (Figs2 and 4, respectively).
A more direct test would be to compare variations in consecutive
push-off and collision phases. The present force data did not include
consecutive steps, and so the proposed effect on redirecting the body
center of mass remains to be tested.
Subjects also appeared to have modified their landing strategy
following heel-strike. As an indicator of such adaptations, we
examined the effective leg length during stance, defined as the
straight-line distance from the sacrum to the calcaneous marker
of the stance foot, normalized to subject leg length. The maximum
effective leg length occurred immediately after heel-strike, and
was reduced by ~2.4% on uneven terrain (mean ±
s.d.=1.140±0.028 for even + foam, 1.113±0.026 for uneven +
foam; P<0.0001). This suggests that subjects adopted a slightly
more crouched posture on uneven terrain, perhaps associated with
increased EMG activity in the thigh muscles. Past research has
suggested that vertical stiffness decreases with a more crouched
posture, for both human running (McMahon et al., 1987) and
walking (Bertram et al., 2002). A more crouched limbed posture
on uneven terrain might also increase compliance and provide a
smoother gait, albeit at higher energetic cost. We also observed
decreased tibialis anterior activation at heel strike, which may be
associated with adaptations for variable conditions at heel-strike.
These overall changes to landing strategy, along with increased
variability in stride period duration, may have contributed to
increased joint work and energetic cost during walking on uneven
terrain.
There are other factors that may have contributed to the increased
energetic cost of walking on uneven terrain compared with even
terrain. Co-activation of muscles about a joint can lead to increased
metabolic cost in human movement (Cavanagh and Kram, 1985).
Although our data suggest an increase in mutual muscle contraction
about the ankle and knee joints (Table2), it is difficult to convert
relative amounts of co-activation to a prediction of energetic cost.
The increased vastus lateralis and vastus medialis activity during
stance (Figs4, 5) could also lead to greater energy expenditure.
Although much of that cost could be quantified by knee power,
production of muscle force may also have an energetic cost beyond
that for muscle work (Dean and Kuo, 2009; Doke and Kuo, 2007).
Although we cannot estimate a cost for co-activation or force
production, it is quite possible that they contributed to the increased
metabolic cost on uneven terrain.
There were several limitations to this study. A limitation of the
data setup was the arrangement of the force plates during over-
ground trials. Force plates placed consecutively would have allowed
us to collect force data during consecutive steps and to analyze
simultaneous work by the leading and trailing legs. Another
limitation was that subjects walked at a controlled walking speed.
This might have constrained their freedom to negotiate terrain by
varying their speed. We also did not test a range of walking speeds
to determine whether uneven terrain causes an altered relationship
between energy cost and speed. We also tested only one pattern and
range of surface heights, with the expectation that greater height
variation would largely have a magnified effect on energetics.
Subjects were also given little time to become accustomed to the
uneven terrain. We had assumed that everyday experience would
allow them to adapt to uneven surface relatively quickly. There was
also reduced ability for subjects to view the terrain surface ahead
of them, because of the limited length of the treadmill. This did not
seem to pose an undue challenge for the small perturbations here,
but we would expect vision to be increasingly important with greater
terrain variations (Patla, 1997).
This study characterizes some of the adaptations that might occur
on uneven terrain. These include relatively minor adaptations in
stepping strategy, increases in muscle activity, and additional work
performed at the hip. A controlled experiment can hardly replicate
the limitless variations of the actual environment, nor can it capture
the entire range of compensations humans might perform in daily
living. But this study does suggest that much of the energetic cost
of walking on uneven terrain may be explained by changes in
0 0.5
1
1.5
2
2.5 3
1
0.8
0.6
0.4
0.2
0
E
.
(W kg
–1
)
Hip
Knee
Ankle
W
.
+
(W kg
–1
)
ΔEff = ≈ 32%
ΔE
.
Even + foam
Uneven + foam
ΔW
.
+
Fig.7. Delta efficiency (ΔEff) for uneven versus even terrain,
defined as the ratio between differences in positive mechanical
power and metabolic power (Δ
W
+
and Δ
E
, respectively; plotted
as filled circles, with units W
kg
–1
). Average joint power is shown
for ankle, knee and hip joints.
THE JOURNAL OF EXPERIMENTAL BIOLOGY

3970
mechanical work from lower limb muscles. As a result, these
findings can potentially influence future designs of robotic
exoskeletons used to assist with locomotion on natural surfaces, as
well as the development of various legged robots. In addition,
numerous studies have been carried out on the biomechanics and
energetics of locomotion in humans and other primates with the
intent of highlighting factors driving the evolution of bipedal
locomotion (Pontzer et al., 2009; Sockol et al., 2007). Our findings
highlight that rather small changes in terrain properties (~2.5cm
terrain height variation) can have substantial impact on muscular
work distribution across the lower limb. Thus, future studies should
take into account how properties of natural terrain, such as terrain
height variability and terrain damping (Lejeune et al., 1998), can
influence potential conclusions relating locomotion biomechanics
and energetics of bipedal evolution.
ACKNOWLEDGEMENTS
The authors thank Sarah Weiss and members of the Human Neuromechanics
Laboratory and Human Biomechanics and Control Laboratory for assistance in
collecting the data.
AUTHOR CONTRIBUTIONS
A.S.V. recruited subjects, managed data collections, completed data analysis and
drafted the manuscript. A.D.K. and M.A.D. provided guidance on experimental
design and helped draft and edit the manuscript. D.P.F. conceived the study,
provided guidance on experimental design, and helped draft and edit the
manuscript. All authors contributed their interpretation of the findings and read and
approved the final manuscript.
COMPETING INTERESTS
No competing interests declared.
FUNDING
This research was supported by a grant from the Army Research Laboratory
[W911NF-09-1-0139 and W911NF-10-2-0022 to D.F.]; Department of Defense
[W81XWH-09-2-0142 to A.K.]; Defense Advanced Research Projects Agency
[Atlas Program to A.K.]; Office of Naval Research [ETOWL to A.K.]; and the
University of Michigan Rackham Graduate Student Fellowship to A.V.
REFERENCES
Bertram, J. E., D’Antonio, P., Pardo, J. and Lee, D. V. (2002). Pace length effects in
human walking: ‘Groucho’ gaits revisited. J. Mot. Behav. 34, 309-318.
Brockway, J. M. (1987). Derivation of formulae used to calculate energy expenditure
in man. Hum. Nutr. Clin. Nutr. 41, 463-471.
Cappellini, G., Ivanenko, Y. P., Dominici, N., Poppele, R. E. and Lacquaniti, F.
(2010). Motor patterns during walking on a slippery walkway. J. Neurophysiol. 103,
746-760.
Cavanagh, P. R. and Kram, R. (1985). Mechanical and muscular factors affecting the
efficiency of human movement. Med. Sci. Sports Exerc. 17, 326-331.
Davies, S. E. and Mackinnon, S. N. (2006). The energetics of walking on sand and
grass at various speeds. Ergonomics 49, 651-660.
Dean, J. C. and Kuo, A. D. (2009). Elastic coupling of limb joints enables faster
bipedal walking. J. R. Soc. Interface 6, 561-573.
Doke, J. and Kuo, A. D. (2007). Energetic cost of producing cyclic muscle force,
rather than work, to swing the human leg. J. Exp. Biol. 210, 2390-2398.
Donelan, M. J., Kram, R. and Arthur, D. K. (2001). Mechanical and metabolic
determinants of the preferred step width in human walking. Proc. Roy. Soc. B 268,
1985.
Donelan, J. M., Shipman, D. W., Kram, R. and Kuo, A. D. (2004). Mechanical and
metabolic requirements for active lateral stabilization in human walking. J. Biomech.
37, 827-835.
Ferris, D. P., Louie, M. and Farley, C. T. (1998). Running in the real world: adjusting
leg stiffness for different surfaces. Proc. Biol. Sci. 265, 989-994.
Ferris, D. P., Liang, K. and Farley, C. T. (1999). Runners adjust leg stiffness for their
first step on a new running surface. J. Biomech. 32, 787-794.
Geyer, H., Seyfarth, A. and Blickhan, R. (2006). Compliant leg behaviour explains
basic dynamics of walking and running. Proc. Biol. Sci. 273, 2861-2867.
Gordon, K. E., Ferris, D. P. and Kuo, A. D. (2009). Metabolic and mechanical energy
costs of reducing vertical center of mass movement during gait. Arch. Phys. Med.
Rehabil. 90, 136-144.
Hak, L., Houdijk, H., Steenbrink, F., Mert, A., van der Wurff, P., Beek, P. J. and
van Dieën, J. H. (2012). Speeding up or slowing down? Gait adaptations to
preserve gait stability in response to balance perturbations. Gait Posture 36, 260-
264.
Kerdok, A. E., Biewener, A. A., McMahon, T. A., Weyand, P. G. and Herr, H. M.
(2002). Energetics and mechanics of human running on surfaces of different
stiffnesses. J. Appl. Physiol. 92, 469-478.
Kuo, A. D. (2002). Energetics of actively powered locomotion using the simplest
walking model. J. Biomech. Eng. 124, 113-120.
Kuo, A. D., Donelan, J. M. and Ruina, A. (2005). Energetic consequences of
walking like an inverted pendulum: step-to-step transitions. Exerc. Sport Sci. Rev.
33, 88-97.
Lejeune, T. M., Willems, P. A. and Heglund, N. C. (1998). Mechanics and energetics
of human locomotion on sand. J. Exp. Biol. 201, 2071-2080.
Margaria, R. (1968). Positive and negative work performances and their efficiencies in
human locomotion. Int. Z. Angew. Physiol. 25, 339-351.
Margaria, R. (1976). Biomechanics and Energetics of Muscular Exercise. Oxford:
Clarendon Press.
Marigold, D. S. and Patla, A. E. (2002). Strategies for dynamic stability during
locomotion on a slippery surface: effects of prior experience and knowledge. J.
Neurophysiol. 88, 339-353.
McAndrew, P. M., Dingwell, J. B. and Wilken, J. M. (2010). Walking variability during
continuous pseudo-random oscillations of the support surface and visual field. J.
Biomech. 43, 1470-1475.
McMahon, T. A., Valiant, G. and Frederick, E. C. (1987). Groucho running. J. Appl.
Physiol. 62, 2326-2337.
Minetti, A. E., Ardigò, L. P. and Saibene, F. (1993). Mechanical determinants of
gradient walking energetics in man. J. Physiol. 472, 725-735.
Murray, M. P., Kory, R. C. and Clarkson, B. H. (1969). Walking patterns in healthy
old men. J. Gerontol. 24, 169-178.
Nakazawa, K., Kawashima, N., Akai, M. and Yano, H. (2004). On the reflex
coactivation of ankle flexor and extensor muscles induced by a sudden drop of
support surface during walking in humans. J. Appl. Physiol. 96, 604-611.
O’Connor, S. M. and Kuo, A. D. (2009). Direction-dependent control of balance
during walking and standing. J. Neurophysiol. 102, 1411-1419.
O’Connor,
S. M., Xu, H. Z. and Kuo, A. D. (2012). Energetic cost of walking with
increased step variability. Gait Posture 36, 102-107.
Pandolf, K. B., Haisman, M. F. and Goldman, R. F. (1976). Metabolic energy
expenditure and terrain coefficients for walking on snow. Ergonomics 19, 683-690.
Patla, A. E. (1997). Understanding the roles of vision in the control of human
locomotion. Gait Posture 5, 54-69.
Pinnington, H. C. and Dawson, B. (2001). The energy cost of running on grass
compared to soft dry beach sand. J. Sci. Med. Sport 4, 416-430.
Pontzer, H., Raichlen, D. A. and Sockol, M. D. (2009). The metabolic cost of walking
in humans, chimpanzees, and early hominins. J. Hum. Evol. 56, 43-54.
Poole, D. C. and Richardson, R. S. (1997). Determinants of oxygen uptake.
Implications for exercise testing. Sports Med. 24, 308-320.
Sawicki, G. S., Lewis, C. L. and Ferris, D. P. (2009). It pays to have a spring in your
step. Exerc. Sport Sci. Rev. 37, 130-138.
Sockol, M. D., Raichlen, D. A. and Pontzer, H. (2007). Chimpanzee locomotor
energetics and the origin of human bipedalism. Proc. Natl. Acad. Sci. USA 104,
12265-12269.
Soule, R. G. and Goldman, R. F. (1972). Terrain coefficients for energy cost
prediction. J. Appl. Physiol. 32, 706-708.
Sponberg, S. and Full, R. J. (2008). Neuromechanical response of musculo-skeletal
structures in cockroaches during rapid running on rough terrain. J. Exp. Biol. 211,
433-446.
Thoroughman, K. A. and Shadmehr, R. (1999). Electromyographic correlates of
learning an internal model of reaching movements. J. Neurosci. 19, 8573-8588.
Wade, C., Redfern, M. S., Andres, R. O. and Breloff, S. P. (2010). Joint kinetics and
muscle activity while walking on ballast. Hum. Factors 52, 560-573.
Weir, J. B. V. (1949). New methods for calculating metabolic rate with special
reference to protein metabolism. J. Physiol. 109, 1-9.
Winter, D. A. and Yack, H. J. (1987). EMG profiles during normal human walking:
stride-to-stride and inter-subject variability. Electroencephalogr. Clin. Neurophysiol.
67, 402-411.
Yang, J. F. and Winter, D. A. (1984). Electromyographic amplitude normalization
methods: improving their sensitivity as diagnostic tools in gait analysis. Arch. Phys.
Med. Rehabil. 65, 517-521.
The Journal of Experimental Biology 216 (21)
THE JOURNAL OF EXPERIMENTAL BIOLOGY