Speed modulation in hylobatid bipedalism: a kinematic analysis.

Abstract

Gibbons are highly arboreal apes, and it is expected that their bipedal locomotion will show some particularities related to the arboreal environment. Previous research has shown that, during hylobatid bipedalism, unsupported phases are rare and stride frequencies are relatively low. This study confirms previous findings, and we suggest that low stride frequencies and the absence of unsupported phases are ways to reduce disadvantageous branch oscillations during arboreal travel. Despite these restrictions, gibbons are able to locomote at a wide range of speeds, implying that they likely exploit other mechanisms to modulate their locomotor speed. To investigate this possibility, we collected video images of a large number of spontaneous bipedal bouts of four untrained white-handed gibbons by using an instrumented walkway with four synchronized cameras. These video images were digitized to obtain a quantification of the 3D kinematics of hylobatid bipedalism. We defined a large number of spatiotemporal and kinematic gait variables, and the relationship between these gait variables and (dimensionless) speed was statistically tested. It was found that gibbons mainly increase stride length to increase their locomotor speed; the main speed-modulating mechanisms are hip and ankle excursion and coupled knee and ankle extension at toe-off. Although aerial phases are rare, gibbons generally adopt a bipedal bouncing gait at most speeds and a clear-cut gait transition, as seen in human locomotion, is absent. Comparison with human and bonobo bipedalism showed that the variability of the 3D joint angles of the hind limb are comparable during human and gibbon bipedalism, and much lower than during bonobo bipedalism. The low variability found in gibbons might be related to constraints imposed by the arboreal environment. These arboreal constraints clearly affect the bipedal gait characteristics of gibbons, but do not constrain the ability to adopt a bipedal bouncing gait during terrestrial locomotion.

Full text

Page 1

Speed modulation in hylobatid bipedalism: A kinematic analysis*
Evie E. Vereeckea,b,*, Kristiaan D’Aou ˆta,c, Peter Aertsa,d
aLaboratorium for Functional Morphology, University of Antwerp, Belgium
bDepartment of Human Anatomy and Cell Biology, University of Liverpool, UK
cCentre for Research and Conservation, Belgium
dDepartment of Movement and Sports Sciences, University of Ghent, Belgium
Received 8 November 2005; accepted 5 July 2006
Abstract
Gibbons are highly arboreal apes, and it is expected that their bipedal locomotion will show some particularities related to the arboreal
environment. Previous research has shown that, during hylobatid bipedalism, unsupported phases are rare and stride frequencies are relatively
low. This study confirms previous findings, and we suggest that low stride frequencies and the absence of unsupported phases are ways to reduce
disadvantageous branch oscillations during arboreal travel. Despite these restrictions, gibbons are able to locomote at a wide range of speeds,
implying that they likely exploit other mechanisms to modulate their locomotor speed. To investigate this possibility, we collected video images
of a large number of spontaneous bipedal bouts of four untrained white-handed gibbons by using an instrumented walkway with four synchro-
nized cameras. These video images were digitized to obtain a quantification of the 3D kinematics of hylobatid bipedalism. We defined a large
number of spatiotemporal and kinematic gait variables, and the relationship between these gait variables and (dimensionless) speed was statis-
tically tested. It was found that gibbons mainly increase stride length to increase their locomotor speed; the main speed-modulating mechanisms
are hip and ankle excursion and coupled knee and ankle extension at toe-off. Although aerial phases are rare, gibbons generally adopt a bipedal
bouncing gait at most speeds and a clear-cut gait transition, as seen in human locomotion, is absent. Comparison with human and bonobo
bipedalism showed that the variability of the 3D joint angles of the hind limb are comparable during human and gibbon bipedalism, and
much lower than during bonobo bipedalism. The low variability found in gibbons might be related to constraints imposed by the arboreal
environment. These arboreal constraints clearly affect the bipedal gait characteristics of gibbons, but do not constrain the ability to adopt a
bipedal bouncing gait during terrestrial locomotion.
? 2006 Elsevier Ltd. All rights reserved.
Keywords: Kinematics; Primate locomotion; Velocity; Joint angles; Gibbon
Introduction
Gibbons live high in the forest canopy (15e35 meters; Git-
tins, 1983), and during arboreal travel, fast brachiation bouts
alternate with short bipedal bouts over horizontal branches,
trunks, and boughs (Ellefson, 1974; Fleagle, 1976; Gittins,
1983; Sati and Alfred, 2002). According to Demes et al.
(1990) and Schmitt (1999), maintaining balance is of major
importance in such an environment, and unsupported phases
within the locomotor cycle, as well as high stride frequencies,
are unfavorable because they might elicit branch oscillations
adverse to stable gaits.
Experimental results on a group of captive white-handed
gibbons (Hylobates lar) seem to confirm the assumptions put
forward by these authors (Vereecke et al., 2006). First, aerial
phases during bipedal locomotor cycles were never observed
during bipedal locomotion over a pole (Vereecke et al.,
2006a) and were very rare on the ground (less than 2% of
all observations; Vereecke et al., 2006a). When aerial phases
are present, duty factors always remain very close to 0.5,
*Grant sponsorship: Research assistant and research project (G.0209.99N)
of the Fund for Scientific Research, Flanders (Belgium).
* Corresponding author. Department of Human Anatomy and Cell Biology,
University of Liverpool, Liverpool L69 3GE, UK.
E-mail address: Evie.Vereecke@liverpool.ac.uk (E.E. Vereecke).
0047-2484/$ - see front matter ? 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jhevol.2006.07.005
Journal of Human Evolution 51 (2006) 513e526

Page 2

suggesting that bipedal strides with prominent aerial phases
are indeed avoided. Moreover, the sudden ‘‘stride-to-stride’’
drop in duty factor that is present at gait transition in human
bipedalismdfrom 0.57 to 0.42 and then further decreasing
to 0.3 at high speeds (Minetti, 1998; Segers et al., in press)d
does not occur. Instead, the duty factor gradually decreases
with speed in order to stagnate at about 0.5.
Second, considering the small size of the gibbons, the abso-
lute stride frequencies are likely high (Heglund and Taylor,
1988; Gatesy and Biewener, 1991). At the average speed of
1.44 m/s on the pole, stride frequency equals 1.49 Hz during
the bipedal locomotion of gibbons (Vereecke et al., 2006a).
According to the concept of Froude numbers for locomotion
(Alexander, 1976; Alexander and Jayes, 1983), this is equiva-
lent to a speed of about 2.5 m/s in humans (slow running, fast
walking), at which stride frequencies are about 1.35 Hz (e.g.,
Cavanagh, 1990; Minetti, 1998). Yet, taking size into account,
i.e., making use of dimensionless frequencies (cf. Alexander
and Jayes, 1983; see below), the frequency in the gibbon
only amounts to about 63% of the stride frequency in humans
at the equivalent speed.
Since excluding an aerial phase in the locomotor cycle
likely constrains stride length, and because locomotor speed
equals the product of stride length and frequency (being rela-
tively low in gibbons; see above), it should follow that the
maximum speed for bipedal locomotion in gibbons remains
limited. However, our observations revealed that the gibbon
specimens in our study group are able to locomote at a remark-
ably wide range of speeds (0.56e2.82 m/s, n¼ 208; Vereecke
et al., 2006a), both arboreally (on a pole) and terrestrially.
According to the principle mentioned above (Froude number),
this is equivalent to speeds of up to 18 km/hr in humans, i.e.,
comparable to very fast distance running! Given the low rela-
tive stride frequencies and the absence of floating to increase
stride length, gibbons must rely on other mechanisms to mod-
ulate stride length in order to attain such high bipedal speed.
During human running, several kinematic factors, such as
hip protraction and retraction, knee and ankle extension (at ini-
tial contact and toe-off), and transverse pelvic rotation (along
the body axis), lead to stride elongation (Wagenaar and Beek,
1992; Novacheck, 1998; LaFiandra et al., 2003), and it can be
expected that gibbons likely exploit similar mechanisms (to
a higher extent) in order to reach their fast bipedal arboreal
performance.
To investigate this, we need detailed 3D kinematics of hy-
lobatid bipedalism. A kinematic study of bipedal locomotion
on the pole was, however, not allowed by zoo protocol, and
it was also practically impossible (outdoor conditions, high-
tech devices, multiple cameras, etc.). However, our previous
study (Vereecke et al., 2006a) showed that gait differences be-
tween pole and level bipedalism were minimal in gibbons and
we can, therefore, use the kinematic study of level bipedalism
to gain insight into the speed-modulating mechanisms of
gibbons.
We quantified the spatiotemporal and kinematic character-
istics of a large number of spontaneous bipedal bouts of four
untrained Hylobates lar subjects, covering a wide range of
speeds. The correlation between these gait variables and speed
was statistically tested using mixed linear models and simple
regression analyses.
Materials and methods
Experimental setup
We videotaped four untrained gibbons (Hylobates lar) dur-
ing spontaneous bipedal locomotion over an instrumentally
monitored walkway in their indoor housing at the Wild Ani-
mal Park Planckendael, Belgium. We used the same setup as
described in Vereecke et al. (2005), which consists of a 4-m-
long walkway with a built-in force and pressure plate, and
which is enclosed by a latticework corridor, through which
the gibbons had to walk when going to their outdoor exhibit
(Fig. 1). Four genlocked S-VHS cameras (50 Hz) surrounded
the setup and recorded the lateral and frontal views. The zoo
protocol did not allow marker setting and interaction with
the animals.
We retained 47 complete gait cycles for analysis without
distinguishing between left and right sides. These sequences
are strictly bipedal bouts (from four subjects) with a speed
range of 0.7e3.5 m/s (Table 1), of which only six bouts had
a duty factor of slightly less then 0.5. Magnitude and
Fig. 1. Experimental setup, consisting of a 4-m-long walkway and four synchronized cameras (AeD).
514
E.E. Vereecke et al. / Journal of Human Evolution 51 (2006) 513e526

Page 3

constancy of the progression speed were assessed from the lin-
ear regression of the horizontal displacement (y-axis) of the
hip against stride duration. The slope of the regression equals
the progression speed and the coefficient of determination, r2,
estimates the constancy of the speed during the bout. For all
retained sequences, the r2values were greater than 0.99.
Digitization
Digitization of the video images was performed manually
with Kwon 3D software (Kwon, 1994). A complete gait cycle
(or stride) was digitized frame by frame in each camera view:
the right side was digitized in views A and B, the left side was
digitized in views C and D, and the head was digitized in every
camera view. All digitizations were executed by the same per-
son using exactly the same procedure to maximize their reli-
ability. We started to digitize from the same frame in each
view, as indicated by a blinking LED (‘‘synchronization
event’’). Twenty points [crown, chin, shoulder (2), elbow
(2), wrist (2), hip (2), knee (2), ankle (2), heel (2), and second
toe (2)] were digitized, resulting in a 14-linked-segment body
model (Fig. 2). Each segment of the model was regarded as
a rigid body with inertial properties that are linked to other
segments at the joints. The model (Fig. 2) consisted of the fol-
lowing segments: head, trunk, left and right upper and lower
arm, left and right hand, left and right upper and lower leg,
and left and right feet, which were defined by digitization of
the 20 points. The raw positional data were filtered using
a 4th Butterworth Low-Pass Filter with 7 Hz cutoff frequency.
The latter was the optimal cutoff frequency, which was deter-
mined using residual-against-frequency plots of the positional
and angular data, as described by Winter (1990: 41e43). The
3D coordinates of each point were recalculated from the four
synchronized planar views (using the 3D-DLT method in
KwonCC 3.01 for Windows; Kwon, 1994). In order to obtain
the transformation parameters, we digitized the eight corners
of a rigid calibration frame of 0.5 ? 1.0 ? 0.5 m3in five con-
secutive frames of each camera view. The reconstruction error
totaled 0.44 cm.
Data analysis
In the gibbon body model, we defined 3D joint angles at the
hip, knee, and ankle, as well as some additional 2D projection
angles, which are illustrated in Figure 3. In the frontal (ZX)
plane, we calculated the trunk tilt, hip tilt, and thigh abduction;
in the transverse plane (XY), we defined foot angle, hip twist
(i.e., maximal ? minimal hip angle) and shoulder-hip twist.
Trunk inclination and thigh angle at initial contact (TIC) and
at toe-off (TTO) were measured in the sagittal (ZY) plane
(Fig. 3). In addition, the linear kinematics of the digitized se-
quences were also used to calculate spatiotemporal parame-
ters. This data set includes an even wider range of speeds
than presented in a previous paper (Vereecke et al., 2005).
The spatiotemporal parameters were defined as follows: stride
length (i.e., Y component of the distance between two consec-
utive contacts of the same foot, measured at the heel), foot
clearance (i.e., Z component of the maximal distance between
the heel and the substrate), step width (i.e., X component of
the distance between left and right heel during the double-
support phase), contact time (i.e., stance-phase duration),
duty factor (i.e., contact time as a fraction of total stride dura-
tion), stride frequency (i.e., inverse of stride duration), and
double-stance period (i.e., duration of the double-support
phase). The V, FC, SW, SL, and SF were scaled to hip height
using the formulae provided by Alexander and Jayes (1983) to
correct for the size differences between the four animals (see
Tables 1 and 2; Gatesy and Biewener, 1991). For each individ-
ual, we calculated the average hip height during the stance
phase, as obtained from the digitization of the corresponding
video images. These individual hip heights varied slightly
between the different sequences, and therefore the hip height
was averaged over the selected sequences.
To allow generation of mean-angle profiles, we resampled
each gait cycle to comprise exactly 26 intervals of 4% of stride
duration (linear interpolation within the boundary-measuring
interval was applied). For each 4% interval, the mean and stan-
dard deviation of all sequences (n ¼ 47) were calculated, and
the results for each joint were presented in a time plot
Table 1
Subject data
SubjectYear of
birth
Sex*Mass
(kg)
Hip height
(m)
Number of
sequences
V*
(m/s)
Ge
Na
Ya
Be
1980
1983
1997
2000
M
F
M
M
7.5
6.5
6.3
3.5
0.338
0.315
0.310
0.267
12
11
11
13
1.03e2.03
0.84e2.27
0.71e3.53
0.94e1.75
*Abbreviations are as follows: F ¼female; M ¼male; V¼ velocity.
Fig. 2. The 14-linked-segment gibbon model with 20 digitization points.
515
E.E. Vereecke et al. / Journal of Human Evolution 51 (2006) 513e526

Page 4

(Fig. 4). To visualize the effect of velocity on the angular ex-
cursions, we classified the bipedal sequences into four speed
classes: ‘‘slow,’’ Froude number <0.5 [or dimensionless
velocity (DV) < 0.7; n¼ 17]; ‘‘moderate,’’ Froude number ¼
0.5e1.0 (DV ¼ 0.7e1.0; n¼ 16); ‘‘fast,’’ Froude number¼
1.0e2.0 (DV ¼ 1.0e1.4; n ¼ 11); and ‘‘sprint,’’ Froude
number> 2.0 (DV > 1.4; n¼ 3); we then calculated average
angular excursions for each speed class. To quantify the vari-
ation of the joint angles within a gait cycle, a coefficient of
variation (CV) of the mean-angle profile was calculated using
the following formula (D’Aou ˆt et al., 2002):
CV ¼
" X
n
i¼1
si=n
!,
qmax?qmin
!#
,100
with n representing the number of intervals (in our case
n¼ 26), i representing the interval number, si representing
the standard deviation at interval i, and qmaxand qminrepre-
senting the maximal and minimal values of the joint angle
found during the mean angle curve. This CV is basically the
mean standard deviation for the mean-angle curve, scaled to
the mean range of movement at the angle considered, and
expressed as a percentage.
Statistical analysis
To assess the relationship between the angular-time profiles
and velocity, some landmark points were defined on the plots:
hip, knee, and ankle angles at initial contact and toe-off; range
of motion at hip, knee, and ankle; timing of the maximal hip,
Fig. 3. Definition of the kinematic variables. Left side: 3D joint angles at hip, knee, and ankle and 2D thigh angle and trunk inclination. Right side: projection
angles in the frontal (ZX) plane [trunk tilt, thigh abduction (Abd), and hip tilt] and in the transverse (XY) plane [hip angle and shoulder-hip twist (SHtwist)].
Table 2
Formulae for calculation of the dimensionless variables
Gait variable Dimensionless variable*
Foot clearance (FC)
Step width (SW)
Stride length (SL)
Velocity (V)
Stride frequency (SF)
Duty factor (DF)
DFC¼ FC/HH
DSW¼SW/HH
DSL¼SL/HH
DV ¼ V=
DSF ¼ SF,
DF¼CT/SD
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðg,HHÞ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p
ðHH=gÞ
p
*Abbreviationsareasfollows:HH¼ hipheight(inmeters);g¼gravitational
acceleration constant¼9.81 m/s2; CT¼contact time (in seconds); SD ¼stride
duration (in seconds).
516
E.E. Vereecke et al. / Journal of Human Evolution 51 (2006) 513e526

Page 5

knee, and ankle extension angles; timing of the maximal knee
and ankle flexion angles; timing of the secondary knee and an-
kle maximal extension angles; and magnitudes of these second-
ary extension angles (relative to the maximal extension angle;
see Fig. 4). The relationship between the kinematic and spatio-
temporal gait parameters and dimensionless velocity was sta-
tistically tested using SAS System 8.02 for Windows (SAS
Institute, 1999). A mixed linear model (SAS MIXED proce-
dure) was used to calculate the linear equations (as a function
of dimensionless velocity) and to account for within-subject
covariation resulting from repeated measures (each animal
made multiple bouts, and observations of the same subject are
usually correlated; Wolfinger and Chang, 1995; Kowalchuck
et al., 2004). The Satterthwaite method (Satterthwaite, 1941)
was used to adjust the degrees of freedom, which is necessary
for repeated measures in unbalanced designs (Kowalchuck
etal.,2004).Plotsofthedifferentkinematicandspatiotemporal
parametersasafunctionofdimensionlessvelocitywere created
with corresponding regression lines. In addition, some simple
regression analyses were performed (proc REG in SAS) to cal-
culate (partial) r2values of regression equations, which was not
possible in the mixed model approach because this procedure
uses restricted maximum likelihood, REML, as the estimation
method. Although these regression analyses do not account
for possible intraindividual correlations (resulting from re-
peated measures), they can be used to illustrate the relative
contribution of some gait characteristics to dimensionless
velocity and stride length. Statistical significance was defined
as p< 0.05 for all analyses.
Results
A complete list of the average values and standard devia-
tions of the collected gait parameters is given in Table 3,
together with the p-values of their correlation with dimension-
less velocity. The average angular excursions of the hip, knee,
and ankle as functions of cycle time are shown in Figure 4, and
minimal and maximal angular excursions of each joint are
given in Table 4. Angular-time profiles (Fig. 5) and angle-angle
plots (Fig. 6) illustrate the effect of speed on the hind-limb joint
movements. In addition, scatter plots (Figs. 7 and 8) are given
to illustrate the significant relationships between the kinematic
gait parameters and dimensionless velocity and to indicate the
variability of the gait parameters.
General pattern and the effect of speed
The knee and ankle are maximally extendeddto 144?and
119?, respectivelydat touchdown and are flexed during the
midstance (Fig. 4, Table 4). There is a second knee extension
in the second half of the stance phase, just prior to hip and an-
kle extension and toe-off. The knee is maximally flexed during
the swing phase (73?). Both knee and ankle are extended twice
during a stride, whereas the hip is only extended once. This
maximal hip extension (151?) coincides with the push-off.
This general pattern is affected by speed, but only a few kine-
matic parameters are found that might contribute to stride
elongation (see below).
Figures 5 and 6 demonstrate the differences in hind-limb
angular excursions between different speed classes: ‘‘slow,’’
Froude number <0.5 (DV< 0.7);
number ¼ 0.5e1.0 (DV ¼0.7e1.0); ‘‘fast,’’ Froude number¼
1.0e2.0 (DV¼ 1.0e1.4); and ‘‘sprint,’’ Froude number > 2.0
(DV > 1.4). The angular-time profiles (Fig. 5) illustrate the
average angular excursions of each speed class as a function
of cycle time, and the angle-angle plots illustrate the speed-
related differences in hip-knee and knee-ankle coordination
and excursions (Fig. 6). Both figures clearly show that hip
‘‘moderate,’’ Froude
Fig. 4. Plot showing the average 3D joint angles at the hip (dotted line), knee (stripes), and ankle (solid line) as a function of gait-cycle time [with confidence
intervals (CI; n¼47)]. Bars at the bottom of the graph indicate the stance phases of the right (gray) and left (black) foot. Vertical lines represent average instant
of toe-off (solid line) with standard deviation (dotted line). Triangles show second knee (KM2) and ankle (AM2) extension.
517
E.E. Vereecke et al. / Journal of Human Evolution 51 (2006) 513e526