Physica A 433 (2015) 7483

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

A synergetic model for human gait transitions

Mohammad Abdolvahab
Center for the Ecological Study of Perception and Action, Department of Psychology, University of Connecticut, 406 Babbidge Rd., Storrs
CT 06269, USA

highlights

A synergetic model for human gait transitions is presented.

Two experiments were conducted to study human gait transitions on a treadmill.
The effect of acceleration and inclination on gait transitions were investigated.
The results of the experiments have been described using the proposed model.

article

info

Article history:
Received 29 November 2014
Received in revised form 13 March 2015
Available online 2 April 2015
Keywords:
Gait transition
Synergetics
Hysteresis
Dynamical systems

abstract
Gait transitions have been considered as bifurcations between states (e.g. walking or running modes) of a nonlinear dynamical system. A top-down synergetic approach to model
gait transitions has been adapted from Frank et al. (2009) and applied to two sets of empirical observations. In this approach, it is assumed that the amplitudes of the spatio-temporal
modes of locomotion satisfy a generic form of evolution equations that are known to hold
for animate and inanimate self-organizing systems. The presented experimental results
focus on hysteresis in human walk-to-run and run-to-walk transitions on a treadmill as a
function of treadmill inclination and acceleration, the rate at which speed was increased or
decreased during experimental trials. The bi-stability in the synergetic model is assumed
to account for the hysteretic transitions. Accordingly, the relevant parameters of the model
were estimated from the empirical data and the models efficacy in predicting the observed
hysteresis effects was evaluated.
2015 Elsevier B.V. All rights reserved.

1. Introduction
Spontaneous switching between behaviors is commonplace. For example, bipeds and quadrupeds switch from walking
to running or vice versa depending on their required locomotion speed. In a laboratory setting, specifically on a treadmill
with varying speed, the transition between gaits (walking or running) has been investigated extensively. It is known that
the transition depends on different parameters such as speed, stride length and stride frequency [1].
The switching between behaviors is based on the perception of an opportunity for action in the environment (e.g.,
running) that one perceives relative to ones dimensions and action capabilities (e.g., leg length, weight, etc.). This idea that
the environment is perceived as body-scaled reflects Gibsons concept of affordances [2]. For example, a treadmill operating
at a particular speed that affords walking for an adult (with longer legs) can only afford running for a child (with shorter
legs). In some studies, it has been considered that a dimensionless ratio, known as a -number [1,3], invariantly specifies
the boundary between walking and running irrespective of different action capabilities of differently sized individuals.

Tel.: +1 860 486 2212; fax: +1 860 486 2760.

E-mail address: mohammad.abdolvahab@uconn.edu.

http://dx.doi.org/10.1016/j.physa.2015.03.049
0378-4371/ 2015 Elsevier B.V. All rights reserved.

M. Abdolvahab / Physica A 433 (2015) 7483

75

In this context it has been proposed that locomotory movements are dynamically similar when their Froude (Fr) numbers
are equal (e.g., Ref. [4]), where Fr = v 2 /(lg ) (i.e., the squared speed, v , divided by the product of leg length, l, and gravitational
acceleration, g). The -number Fr can be characterized as the ratio of inertial to gravitational forces. In what follows, the Fr
number at which the transition occurs will be the critical Fr number.
The physical interpretation of critical Fr is that the walk-to-run transition occurs when the acceleration of the bodys
center of gravity resulting from inertial forces exceeds the acceleration caused by gravity. In several studies, it has been
shown that the transition as measured by the critical Froude number differs experimentally for systematically increasing
and systematically decreasing the -number or the speed: there is hysteresis (e.g. Refs. [1,57]). Hysteresis is defined as a
property of a system such that an output value is not a strict function of the corresponding input, but also incorporates some
delay or history dependence. In particular, hysteresis refers to the case when the response for a decrease in the input variable
is different from the response for an increase. Hysteresis has been identified in various psychological phenomena and in
particular in affordance research (e.g. in perceptual speech categorization in Tuller et al. [8]; in perceptual judgments of the
maximum slope of a surface that afforded upright posture with either visual or haptic exploration in Fitzpatrick et al. [9]; in
perception of maximum sit-on-ability in Hirose et al. [10]; in perception of grasp-ability of objects in Richardson et al. [11]).
The hysteresis phenomenon in gait transition higher transition speed for walk-to-run (WR) than for runwalk (RW) has
also been reported in a number studies (e.g. Refs. [1,7]).
The aim of the present work is to study the effects of biomechanical constraints on hysteresis using a synergetic dynamical model. To date, there is very little effort in mathematical modeling of gait transition (for an exception see e.g. the
probabilistic model by Li [12]). The dynamical model used in this study is adapted from the grasping transition (GT) model
in Frank et al. [13]. The GT model [13] was originally derived from a model proposed by Haken [14]. Hakens model was
aimed to describe pattern recognition and oscillatory perceptual processes of ambiguous patterns [15]. The model proposed
by Haken, which has often been interpreted as a neural network model (see Refs. [16,17]), assumes that attractors and
repellers of the neural system of humans and animals determine states of neural activity. In the context of behavioral transitions, these are the attractors and repellers of the organismenvironment system that determine states of behavior. For
instance, in grasping transition behavior, the information about the to-be-grasped objects (i.e. object size) and the relevant
properties of the individuals action system (i.e. hand span) constrains and guides the grasping behavior of an individual.
1.1. Synergetic model
Historically, gait transitions are considered to correspond to bifurcations between the states of a nonlinear dynamical
system (e.g. Refs. [1821]). The behavior of such a system can be described in terms of an order parameter, which can be
the relative phase between limbs or limb segments. Continuous variation of a control parameter, e.g. treadmills speed, can
induce bifurcations in the order parameter [20].
The mathematical theory of synergetics assumes that the behavior of a system is decomposable into the time-evolution of
its dynamical modes [14]. A systems dynamical mode is characterized by an eigenvector and an eigenvalue. An eigenvector
corresponds to an observed mode, behavior, or pattern. An eigenvalue describes the rate of change of a dynamical mode.
Note that we will re-interpret below the eigenvalue as availability parameter.
In the context of human WR or RW gait transitions, we assume 1 (t ) (or 1 ) and 2 (t ) (or 2 ) represent the dynamical
evolution of generalized amplitudes of the walking and running modes, respectively.1 Then, as in Frank et al. [13], we assume
the following potential function for the system with the two dominant modes:
V (1 , 2 ) =

b 2 2 c 2
2
1 12 + 2 22 +
1 2 +
1 + 22 .

(1)
2
4
When 1 > 0 and 2 = 0, walking (YW ) is performed and when 2 > 0, 1 = 0, running (YR ) is performed. In Eq. (1), 1 and
2 are availability parameters defining the possibilities for walking and running, respectively. The magnitude of determines
how strongly one of the modes is activated. A behavioral mode is available for > 0 and is not available for < 0.2 Fig. 1
depicts the potential function with constant coefficients b and c. For 1 > 0 and 2 < 0, the potential shows minima on
2 = 0 axis indicating that the walking mode is the only stable mode (Fig. 1; left panel). For 1 < 0 and 2 > 0, the potential
includes minima on 1 = 0 axis explaining that the running mode is the stable mode (Fig. 1; center panel). For 1 and 2 > 0,
depending on the boundary conditions at [1 (0), 2 (0)], both walking and running modes can be stable (Fig. 1; right panel).
Dynamical evolution of amplitudes of the systems modes is governed by the following equations:
d
dt
d
dt

1 =

V (1 , 2 )
1

(2)

2 =

V (1 , 2 )
.
2

(3)

1 Note that spatial characteristic of the modes can remain non-specified in this approach. In other words, magnitudes of and are not directly
1
2
evaluated.
2 Although for > 0 a mode is available, the mode can be either stable or unstable. Only stable available modes are performed. In general, if (or )

is much larger than 2 (1 ), then YW (or YR ) is stable. For 1 = 2 , both modes are stable.

76

M. Abdolvahab / Physica A 433 (2015) 7483

Fig. 1. Left panel: potential function for 1 > 0 and 2 < 0, representing mono-stability in which only the walking mode is stable; Center panel: potential
function for 1 < 0 and 2 > 0, showing mono-stability for the running mode; Right panel: potential function for 1 and 2 > 0, representing bi-stability;
both walking or running modes can be active depending on the initial conditions.

From Eqs. (1) to (3), it can be followed:

d
dt
d
dt

1 = 1 1 (b + c )22 1 c 13

(4)

2 = 2 2 (b + c )12 2 c 23 .

(5)

By dividing the coefficient of mixed terms in Eqs. (4) and (5), 22 1 and 12 2 , by c, we define a new parameter g (=1 + b/c),
the interaction parameter, which represents the strength of the competition between walking and running modes in a
winner-takes-all race. The stronger the competition between the modes, the larger is the value of g.
As reviewed above, on the one hand, 1 and 2 determine the stability of the behavioral modes, walk or run, respectively.
On the other hand, in laboratory experiments, the locomotion speed and consequently the Froude number Fr is manipulated.
Therefore, it is assumed that 1 and 2 depend on Fr. Without loss of generality, it can be assumed that the availability
parameter or is a linear function of the control parameter Fr:

1 = L1,0 Fr

(6)

2 = L2,0 + Fr .

(7)

In the GT model [13], Eqs. (6) and (7), L1,0 and L2,0 are constants. Given that the stability of walking or running modes is
primarily determined by the availability parameters 1 and 2 , if an availability parameter is large, the corresponding mode
is attractivethat is, there is a strong tendency to perform the corresponding mode. According to Eqs. (6) and (7), walking
mode becomes less attractive as Fr increases, whereas running mode becomes more attractive as Fr increases. It can be
shown that at a critical Froude number where 2 = g 1 , a transition from walking to running can be observed [13]. Solving
1 = 1 Fr , 2 = L2,0 + Fr and 2 = g 1 for Fr, one obtains the critical value Fr c ,2 for ascending sequence in the form:
Fr c ,2 =

g L 2 ,0
1+g

(8)

In a similar manner, one can derive Fr c ,1 for descending sequence:

Fr c ,1 =

1 g L2,0
1+g

(9)

Combining Eqs. (8) and (9), one can compute 1Fr as:

(g 1) 1 + L2,0
1Fr =
.
1+g

(10)

When there is positive hysteresis in the system (1Fr > 0), the system is bi-stable. The bi-stability in the system appears
when g > 1. The larger the g is, the larger the competition between the modes and hence the wider is the bi-stable
domain [22]. Accordingly, the larger parameter g is, the more positive is the hysteresis in the system. The offset saturation
value L2,0 is estimated as:
L2,0 = 1 Fr c ,1 Fr c ,2 .

(11)

When substituting Eq. (11) in Eq. (10) and solving for g, one obtains:
2 Fr c ,2 + Fr c ,1 + 1Fr

g =

2 Fr c ,2 + Fr c ,1 1Fr

Eqs. (11) and (12) are the estimators for the model parameters g and L2,0 .

(12)

M. Abdolvahab / Physica A 433 (2015) 7483

77

Fig. 2. Schematic prediction about the effect of inclination on the hysteresis in gait transition; with increasing inclination, the critical Froude numbers are
shifted toward lower values.

1.2. Interpretation of model parameters

The parameter g reflects the strength of the competition between the behavioral modes. When g increases, the hysteresis
becomes more positive and if there is positive hysteresis then the hysteretic domain (i.e. the hysteresis size) becomes larger.
The parameter L2,0 is a measure for the overall strength of the attractor of the second mode with respect to the first mode.
When L2,0 increases then the attractor of the second mode becomes stronger in the sense that its basin of attraction becomes
larger. This implies that the critical Froude numbers at which modemode transitions happen are shifted to lower values.
2. Experimental methods
Dynamical systems theory views a specific gait pattern as an emergent behavior that arises from the collective behavior
of all contributing sub-systems, including both the musculoskeletal and central nervous systems [1]. Other constraints, such
as those dictated by the environment (e.g., gravitational forces, slippery walking surfaces) and the task itself (e.g., walking
or running at different velocities and/or different stride frequencies) also contribute to shaping the behavior of the system.
In what follows, two experiments are presented in which two biomechanical parameters (acceleration and inclination) are
manipulated to assess the influence of biomechanical coupling on hysteresis direction.
Experiment 1 was aimed at partially re-examining the experimental results presented by Li [12]. His study, using a
continuous protocol,3 showed that acceleration (as one of many biomechanical constraints at work) influenced the WR
and RW probability landscape differently. High acceleration magnitude produced a higher WR transition speed than RW
(positive hysteresis). When the acceleration decreased, the amount of hysteresis also decreased. For the lowest acceleration,
the positive hysteresis reversed to negative hysteresis. In the present experiment, using a plateau protocol,4 a treadmill with
varying speed at two constant acceleration magnitudes, one equivalent to the lowest acceleration in Li [12] and one at a lower
rate, was used to determine the transition speed.
In Experiment 2, the effect of another biomechanical constraint on the dynamics of gait transition was examined. It has
been clearly demonstrated that the transition speed decreases with increasing treadmill inclination [7,25]. Fig. 2 demonstrates schematically the prediction about the change in the critical Froude numbers with the treadmills inclination. Although it has been suggested that treadmill inclination may also have an effect on the amount of hysteresis, the evidence is
inconclusive [23]. Hreljac [7,23] measured both walk-to-run (WR) and run-to-walk (RW) for several inclination conditions,
finding no difference in the amount of hysteresis between different inclination conditions. Diedrich et al. [25], however,
suggested that increasing inclination might decrease the amount of hysteresis. Experiment 2 was aimed at examining how
changing treadmills inclination affects the dynamics of gait transition and how the estimated model parameters based on
these empirical results fit with the predictions.
2.1. Experiment 1
Two groups of participants underwent experimental trials for each of the tested acceleration or deceleration magnitudes.
The first group comprised 18 participants (2 women, 16 men; mean age = 19.3, SD = 0.8; mean leg length = 94.0 cm) from
the University of Connecticut. The subjects participated in the experiment for partial fulfillment of a course requirement. The
second group also comprised 18 participants (9 women, 9 men; mean age = 21.7, SD = 2.4; mean leg length = 89.2 cm) from
the University of Connecticut. The subjects in this group were paid at a rate of $10 per hour of participation. The universitys
Institutional Review Board approved all procedures. A Trackmaster TMX-425 treadmill (180 kg capacity, running surface of
76 160 cm, speed range 0.225.36 m/s, inclination range 0%25%) with manual controller and standard safety siderails
was used.
3 In the continuous protocol, the speed of the treadmill is increased or decreased continuously using a constant acceleration or deceleration [23,24]. In
this manner, the participant experiences a constantly varying speed during the trial.
4 In the plateau protocol, the speed is increased or decreased by constant increments at constant intervals. Accordingly, in each trial, the participant
walks or runs at constant speed for certain amount of time based on the testing acceleration or deceleration.

78

M. Abdolvahab / Physica A 433 (2015) 7483

The first group of subjects was tested on the treadmill with constantly increasing or decreasing speed. The speed was
manually increased or decreased by the experimenter at constant increments of 0.09 m/s every 5 s (yielding a constant
plateau acceleration or deceleration of approximately 0.02 m/s2 ). The second group was tested at constant increments
of 0.09 m/s every 10 s (yielding a constant plateau acceleration or deceleration of approximately 0.01 m/s2 ). In the first
group, each participant completed four Ascending sequences (or WR) and four Descending sequences (or RW) of treadmills
speeds. The transition speeds for this group were based on averaging across repetitions. In the second group, each participant
completed only one Ascending sequence and one Descending sequence of treadmills speeds. Participants were told which
sequence (ascending or descending) they were performing at the beginning of a trial. Before the experimental trials,
participants completed a practice trial for each of the sequences. The participants were asked to avoid holding safety siderails
unless necessary. The performance sequence (WR or RW first) was counterbalanced across participants. The subjects leg
length was defined by adding the measure of hip-to-knee and knee-to-ankle. The experimental speed range for the first
group was between 1.34 and 2.68 m/s (3.06.0 mph). The experimental speed range for the second group was between 1.5
and 2.5 m/s (3.45.6 mph).
The experiment was based on a 2 (acceleration: 0.01 m/s2 , 0.02 m/s2 ) 2 (sequence: Ascending, Descending) betweensubjects design. The critical Froude numbers were computed for each participant by the respective transition speed squared,
divided by leg length and gravitational acceleration (g, equal to 9.81 m/s2 ). The Fr c ,1 and Fr c ,2 (critical Fr for descending
and ascending sequences, respectively) values were then substituted into Eqs. (11) and (12) to calculate the values of L2,0
and g, respectively, for different conditions.
2.2. Experiment 2
A group of 15 subjects (5 women, 10 men; mean age = 20.3, SD = 1.8; mean leg length = 83.8 cm) from the University of
Connecticut participated in the experiment. The subjects were paid at a rate of $10 per hour of participation. The universitys
Institutional Review Board approved all procedures. The same Trackmaster TMX-425 treadmill (180 kg capacity, running
surface of 76 160 cm, speed range 0.225.36 m/s, inclination range 0%25%) with manual controller as in the Experiment
1 was used in this experiment.
Each participant completed two repetitions of Ascending sequences (WR) and two repetitions of Descending sequence
(RW) of treadmills speeds for 3 different inclinations (0, 5% and 10%,5 manually adjusted by the experimenter). Participants
were told which sequence (ascending or descending) they were performing at the beginning of a trial. However, they were
not explicitly told that the inclination was being manipulated. Before the experimental trials, participants completed a
practice trial for each of the sequences at zero inclination. The final results were based on averaging across repetitions. The
participants were asked to avoid holding safety siderails unless necessary. The performance sequence (WR or RW first) and
inclination conditions were counterbalanced across participants. The subjects leg length was defined by adding the measure
of hip-to-knee and knee-to-ankle. The subjects were run on the treadmill with constantly increasing or decreasing speed at
constant increments of 0.045 m/s every 2 s (yielding a constant acceleration or deceleration of approximately 0.02 m/s2 ).
The experimental speed range was between 1.5 and 2.5 m/s.
The experiment was based on a 3 (inclination: zero inclination, 5% inclination, 10% inclination) 2 (sequence: Ascending,
Descending) within-subjects design. There were two repetitions for each trial. Hence, participants performed twelve trials
in sum. The transition speeds were averaged across repetitions. The critical Froude numbers were computed for each
participant as in Experiment 1. The Fr c ,1 and Fr c ,2 (critical Fr for descending and ascending sequences, respectively) values
for the three inclinations were then substituted into Eqs. (11) and (12) to calculate the values of L2,0 and g parameters for
different inclinations.
3. Results and discussion
3.1. Experiment 1
The purpose of this experiment was to determine whether acceleration, as a biomechanical constraint on human gait
transitions, could significantly affect the dynamics of behavioral transitions. In particular, it was sought to examine whether
the amount of (positive) hysteresis can be diminished toward a critical point transition (where the speed for ascending
and descending trials would be approximately equal) or negative hysteresis. The reduction of hysteresis toward zero or a
negative value, expected to result in significantly smaller values of parameter g. Conversely, it was expected that L2,0 would
increase.
In the ascending sequences, participants switched from walking to running at larger Froude numbers than in the
descending sequences; overall, positive hysteresis was exhibited (see Table 1 for individual behavioral transitions for the
two acceleration conditions). Also Fig. 3 demonstrates mean critical Froude numbers for each condition.

5 The angle of the slope of the treadmill can be calculated from the percent inclination by the inverse tangent of the percent inclination divided by 100.
For example, to find the angle of inclination of 5%, one can compute the inverse tangent of 0.05, which turns out to be about 2.9. Similarly, the angle of
inclination of 10% is about 5.7.

M. Abdolvahab / Physica A 433 (2015) 7483

79

Table 1
Distribution of behavior modes for participants in Experiment 1.
Acceleration

Positive hysteresis
Negative hysteresis
Critical point

Ascending

0.02

0.01

17
1
0

13
2
3

Descending

Mean Critical Froude Number

0.70
0.60
0.50
0.40
0.30
0.20
0.10

Acceleration = 0.02 m/s2

Acceleration = 0.01 m/s2

Fig. 3. Mean critical Froude numbers for walk-to-run (WR) or run-to-walk (RW) transitions for different accelerations or decelerations (rates at which the
speed was increased or decreased, respectively, during an experimental trial).

Fig. 4. The classical (positive) hysteresis for different Acceleration conditions: participants switched at larger Froude numbers for ascending trials than for
descending trials, the amount of hysteresis or the area between ascending and descending curve did not significantly change across the two accelerations.

Fig. 4 shows the percentage of participants running as a function of the Froude number for the two acceleration rates and
the two sequences for each, ascending and descending. A 2 (Acceleration) 2 (Sequence) between-subjects ANOVA on the
mean critical Froude numbers revealed that there was a significant effect of Sequence, F (1, 34) = 40.71, p < 0.001.6 In the
ascending sequences, participants switched from walking to running at larger Froude numbers (Mean = 0.45; SD = 0.08)
than the descending sequences (Mean = 0.39; SD = 0.08). However, the interaction effect between Acceleration and
Sequence was not significant; the amount of hysteresis did not significantly change across the two different accelerations,
1Fr %0.02 m/s2 = 0.069 and 1Fr %0.01 m/s2 = 0.054. These results demonstrate that changing the rate at which the speed

6 A 5% significance level for hypothesis testing was used for all analyses.

80

M. Abdolvahab / Physica A 433 (2015) 7483

Ascending

Descending

Mean Critical Froude Number

0.60
0.50
0.40
0.30
0.20
0.10

%5

%10

Inclination

Fig. 5. Mean critical Froude numbers for WR or RW transitions for different inclinations.

changes during an experimental trial do not significantly affect the amount of hysteresis. The hysteresis remained positive
across the two acceleration rates.
The model parameters g and L2,0 were estimated for the two acceleration conditions using Eqs. (11) and (12). As explained
earlier, parameter g represents the interaction among stable modes of the system, namely walking and running.
If the amount of hysteresis would be affected by acceleration, it would be expected that parameter g significantly changes
across the two different acceleration conditions. An ANOVA showed that the effect of Acceleration for g was not significant.
The values of g decreased minimally when the acceleration rate decreased (Mean0.02 m/s2 = 1.32, SD0.02 m/s2 = 0.29;
Mean0.01 m/s2 = 1.29, SD0.01 m/s2 = 0.29).
With regard to the L2,0 parameter, if hysteresis was significantly different across acceleration conditions, it would be
expected to observe a significant effect of Acceleration on L2,0 . An ANOVA on the L2,0 values for different accelerations
yielded no significant effect. L2,0 was not significantly different from one acceleration condition to another (Mean0.02 m/s2 =
0.17, SD0.02 m/s2 = 0.15; Mean0.01 m/s2 = 0.14, SD0.01 m/s2 = 0.16).
3.2. Experiment 2
The purpose of this experiment was to examine whether by changing the treadmills inclination, as another
biomechanical constraint on human gait transitions, one could observe a significant change in the amount of hysteresis.
Similar to the previous experiment, model parameter estimated from empirical observation can verify the hypothesis with
regard to the characteristics of gait transition dynamics under this biomechanical constraint. The reduction of hysteresis
toward zero or a negative value is associated with decreasing the value of g, indicating reduced competition between the
behavioral modes.
The experimental results showed that participants exhibited positive hysteresis. In the Ascending sequences, participants
switched from walking to running at larger Froude numbers (Mean = 0.46; SD = 0.06) than the Descending sequences
(Mean = 0.36; SD = 0.07). Fig. 5 demonstrates mean critical Froude numbers for all conditions.
Fig. 6 shows the percentage of participants running as a function of the Froude number for the three inclination conditions
and the two sequences for each, ascending and descending. The figure depicts positive hysteresis for all three conditions.
A 3 (Inclination) 2 (Sequence) within-subjects ANOVA on the mean critical Froude numbers revealed that there was
a significant effect for Inclination; F (2, 28) = 42.14, p < 0.001 and a significant effect for Sequence (Ascending vs.
Descending); F (1, 14) = 86.56, p < 0.001. However, the interaction effect between Sequence Inclination was not
significant replicating the results of Hreljac et al. [23]; participants did not show significant difference in the amount of
hysteresis for different inclination conditions (Mean 1Fr 0% = 0.12, Mean 1Fr 5% = 0.10 and Mean 1Fr 10% = 0.09).
These results indicate that increasing levels of inclination significantly moved the attractor location, however, it did not
significantly change the amount of hysteresis.
With regard to the interaction parameter g in the model, if the amount of hysteresis would be affected across different
inclination levels, it would be expected that g significantly changes across conditions due to the larger or smaller bi-stable
zone. An ANOVA showed only a marginally significant effect of Inclination for g, F (2, 28) = 2.63, p = 0.09. The values of
g decreased mildly with the inclination level (Mean0% = 1.59, SD0% = 0.24; Mean5% = 1.42, SD5% = 0.29; Mean10% =
1.42, SD10% = 0.27).
With regard to the L2,0 parameter, it was expected that it would be larger for the higher inclination level. This is due
to the fact that participants switched to running mode earlier. Hence, the prediction would be the larger availability of
this stable state. An ANOVA on the L2,0 values for the different inclination levels yielded a significant effect of Inclination,
F (2, 28) = 46.13, p < 0.001. L2,0 was larger for inclination levels different from zero (Mean10% = 0.26, SD10% =
0.12; Mean5% = 0.18, SD5% = 0.13) than for the zero inclination condition (Mean0% = 0.10, SD0% = 0.14).
The results of Experiment 2 indicate that even though changing the inclination of the treadmill, as another biomechanical
constraint on the system, pushes critical Froude numbers toward lower values, the positivity of the hysteresis remains stable.

M. Abdolvahab / Physica A 433 (2015) 7483

81

Fig. 6. The hysteresis for different inclinations, the amount of hysteresis did not significantly change across Inclination conditions.

In other words, the biomechanical constraint shifts the attractors to the lower values; however, it is always the case that
the WR critical Froude number is larger than that for the RW. Also, according to the ANOVA results, the non-significant
interaction effect between Inclination and Sequence indicates that the amount of hysteresis does not significantly change
across different inclinations.
4. General discussion
In this study, two experiments examined the effect of biomechanical constraints on the behavioral dynamics of gait
transitions. The model parameters of the adapted synergetic model from Ref. [13] for human gait transition were estimated
based on the empirical results of the two experiments. Experiment 1 that was in the first place inspired by the empirical
findings in Li [12], studied the effect of acceleration on the hysteresis in gait transition. According to Lis findings, acceleration
(or deceleration) or the rate at which the speed of the treadmill is changed during an experimental trial can affect the
amount and the direction of hysteresis. He observed that among five different tested acceleration rates, using a continuous
protocol, the lowest tested acceleration changed the direction of hysteresis from positive hysteresis (higher transition speed
for WR transition than for RW transition) to negative hysteresis (higher transition speed for RW transition than for WR
transition). In Experiment 1, an equivalent acceleration rate of lowest tested acceleration by Li, using a plateau protocol,
did not yield similar results; positive hysteresis was generally observed. The observed positive hysteresis was also stable
for a lower acceleration rate. The statistical analysis showed that the amount of hysteresis did not significantly change
across the two acceleration rates. One possible explanation for the discrepancy between the present results and those of
Li can be the difference in the adopted protocol. In fact, this is not uncommon in the available literature on gait transition;
there is considerable variability among the results across various studies using different techniques and methodologies [26
29]. From a biomechanical point of view, the WR transition has been hypothesized (e.g. in Refs. [7,30,31]) to be triggered in

82

M. Abdolvahab / Physica A 433 (2015) 7483

response to stress in the dorsiflexor muscles as walking speed approaches the preferred transition speed. This stress is likely
to be apparent for subjects when given a reasonably long decision period, as is done during a plateau trial [23]. The more
acute situation of the continuous protocol may not allow subjects to perceive dorsiflexor stress as easily. Hence, it can be
argued that a plateau protocol is more accurate for determining the transition speed than a continuous protocol. However,
in the Experiment 1 in the present study, the number of observed critical point and negative hysteresis increased when the
acceleration was decreased. This tendency to earlier switching can be attributed to the effects of fatigue due to the slower
rate of completion of a trial. From the modeling perspective, the statistical results showed that the interaction parameter b
did not significantly differed across conditions explaining that the bi-stability in the system remained in the same range (no
significant tendency toward earlier switches or negative hysteresis). Also, the L2,0 parameter was not significantly affected
by the acceleration indicating the stability in the availability of YR .
In Experiment 2, another biomechanical constraint namely inclination was manipulated to examine its effect on the
dynamics of gait transition. In this experiment, participants also showed positive hysteresis in all conditions. The critical
Froude numbers significantly decreased with higher level of inclination. However, the amount of positive hysteresis
was unaffected by the inclination. The implications of these results was similar to those of Experiment 1 that is the
biomechanical constraint of inclination did not reverse the direction of hysteresis from positive to negative and the amount
of the hysteresis did not differ significantly across conditions. Moreover, estimated model parameters confirmed this
reasoning. The interaction parameter g was not significantly affected by the inclination level meaning that the bi-stability
remained largely unaffected by inclination. On other hand, L2,0 parameter significantly increased with the inclination level
demonstrating significant change in the attractor locations. In fact, participant switched from walking to running earlier at
higher inclinations and hence larger availability of YR .
In sum, as discussed, the impact of two biomechanical constraints on the dynamics of gait transitions was studied though
experimental observation and modeling. The analysis of empirical results and estimated model parameters confirmed the
predictions about the manipulation of the dynamical layout of the attractive states in bi-stable gait transition behavior. The
present results together with the verified model predications support the notion that the principles of self-organization
apply to human gait transitions.
Acknowledgments
This work was conducted as part of the doctoral dissertation of the author while at the Center for the Ecological Study of
Perception and Action at the University of Connecticut. The author is thankful to Professor Michael Turvey for his guidance
in conducting the study and help in preparing the present manuscript. The study was partially supported by NSF Grant
BCS-1344725 (INSPIRE Track 1).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]

F.J. Diedrich, W.H. Warren, Why change gaits? Dynamics of the walk-run transition, J. Exp. Psychol. Hum. Percept. Perform. 21 (1995) 183202.
J.J. Gibson, The Ecological Approach to Visual Perception, Houghton Mifflin, Boston, 1979.
W.H. Warren, Perceiving affordances: visual guidance of stair climbing, J. Exp. Psychol. Hum. Percept. Perform. 10 (1984) 683703.
R.M. Alexander, Exploring Biomechanics: Animals in Motion, W. H. Freeman, New York, 1992.
M. Abdolvahab, Positive hysteresis and negative hysteresis in human gait transition (Doctoral dissertation), University of Connecticut, 2014.
A. Hreljac, Preferred and energetically optimal gait transition speeds in human locomotion, Med. Sci. Sports Exerc. 25 (1993) 11581162.
A. Hreljac, Effects of physical characteristics on the gait transition speed during human locomotion, Hum. Mov. Sci. 14 (1995) 205216.
B. Tuller, P. Case, M. Ding, J.A. Kelso, The nonlinear dynamics of speech categorization, J. Exp. Psychol. Hum. Percept. Perform. 20 (1994) 316.
P. Fitzpatrick, C. Carello, R.C. Schmidt, D. Corey, Haptic and visual perception of an affordance for upright posture, Ecol. Psychol. 6 (1994) 265287.
N. Hirose, A. Nishio, The process of adaptation to perceiving new action capabilities, Ecol. Psychol. 13 (2001) 4969.
M.J. Richardson, K.L. Marsh, R.M. Baron, Judging and actualizing intrapersonal and interpersonal affordances, J. Exp. Psychol. Hum. Percept. Perform.
33 (2007) 845859.
L. Li, Stability landscapes of walking and running near gait transition speed, J. Appl. Biomed. 16 (2000) 428435.
T.D. Frank, M.J. Richardson, S.M. Lopresti-Goodman, M.T. Turvey, Order parameter dynamics of body-scaled hysteresis and mode transitions in
grasping behavior, J. Biol. Phys. 35 (2009) 127147.
H. Haken, Synergetic Computers and CognitionA TopDown Approach to Neural Nets, Springer-Verlag, Berlin, 1991.
T. Ditzinger, H. Haken, Oscillations in the perception of ambiguous patterns a model based on synergetics, Biol. Cybernet. 61 (1989) 279287.
P.C. Bressloff, Neural networks, lattice instantons, and the anti-integrable limit, Phys. Rev. Lett. 75 (1995) 962965.
P.C. Bressloff, A self-organizing network in the weak-coupling limit, Physica D 110 (1997) 195208.
J.J. Collins, I. Stewart, Hexapodal gaits and coupled nonlinear oscillator models, Biol. Cybernet. 68 (1993) 287298.
G. Schner, W.Y. Jiang, J.S. Kelso, A synergetic theory of quadrupedal gaits and gait transitions, J. Theoret. Biol. 142 (1990) 359391.
H. Haken, J.S. Kelso, H. Bunz, A theoretical model of phase transitions in human hand movements, Biol. Cybernet. 51 (1985) 347356.
G. Taga, Y. Yamaguchi, H. Shimizu, Self-organized control of bipedal locomotion by neural oscillators in unpredictable environment, Biol. Cybernet.
65 (1991) 147159.
S.M. Lopresti-Goodman, M.T. Turvey, T.D. Frank, Behavioral dynamics of the affordance graspable, Atten. Percept. Psychophys. 73 (2011) 19481965.
A. Hreljac, R. Imamura, R.F. Escamilla, W.B. Edwards, Effects of changing protocol, grade, and direction on the preferred gait transition speed during
human locomotion, Gait Posture 25 (2007) 419424.
L. Li, J. Hamill, Characteristics of the vertical ground reaction force component prior to gait transition, Res. Q. Exerc. Sport 73 (2002) 229237.
F.J. Diedrich, W.H. Warren, The dynamics of gait transitions: Effects of grade and load, J. Mot. Behav. 30 (1998) 6078.
A.J. Hanna, A.B. Abernethy, R.J. Neal, R.J. Burgess-Limerick, Triggers for the transition between human walking and running, in: W.A. Sparrow (Ed.),
Energetics of Human Activity, Human Kinetics, Champaign, 2000, pp. 124164.
R. Kram, A. Domingo, D.P. Ferris, Effect of reduced gravity on the preferred walk-run transition speed, J. Exp. Biol. 200 (1997) 821826.
A.J. Raynor, C.J. Yi, A.B. Abernethy, Q.J. Jong, Are transitions in human gait determined by mechanical, kinetic or energetic factors? Hum. Mov. Sci. 21
(2002) 785805.

M. Abdolvahab / Physica A 433 (2015) 7483

83

[29] M.T. Turvey, K.G. Holt, M.E. LaFiandra, S.T. Fonseca, Can the transitions to and from running and the metabolic cost of running be determined from
the kinetic energy of running? J. Mot. Behav. 31 (1999) 265278.
[30] A. Hreljac, A. Arata, R. Ferber, J.A. Mercer, B.S. Row, An electromyographical analysis of the role of dorsiflexors on the gait transition during human
locomotion, J. Appl. Biomed. 17 (2001) 287296.
[31] B.I. Prilutsky, R.J. Gregor, Swing-and support-related muscle actions differentially trigger human walkrun and runwalk transitions, J. Exp. Biol. 204
(2001) 22772287.