Simulation Modelling Practice and Theory 40 (2014) 132–143

Contents lists available at ScienceDirect

Simulation Modelling Practice and Theory

journal homepage: www.elsevier.com/locate/simpat

Modelling of a robotic leg using bond graphs

Victor Ragusila, M. Reza Emami ⇑
University of Toronto Institute for Aerospace Studies, Toronto, Canada

a r t i c l e i n f o a b s t r a c t

Article history: This paper studies the bond graph model of a robotic leg mechanism, and discusses meth-
Received 14 February 2013 ods of extracting significant features of system dynamics through simpler models. The goal
Received in revised form 26 June 2013 is to determine a set of simpler mechanisms with similar dynamic behaviour to that of the
Accepted 18 September 2013
original leg in various phases of its motion. The paper is divided in two sections. In the first
Available online 19 October 2013
section, a modular bond-graph representation of the leg mechanism is determined. In the
second section, two algorithms are applied to simplify the bond graph representation. The
Keywords:
first algorithm determines the relevant dynamic elements of the system for each phase of
Bond graph
Model simplification
motion, and the second algorithm finds the simple mechanism described by the remaining
Legged locomotion dynamic elements. In addition to greatly simplifying the control system of the robotic leg,
using simpler mechanisms with similar behaviour provides a greater insight into the
dynamics of the system.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction

Legged robots offer a number of advantages compared with their wheeled counterparts. They are able to navigate rough
terrain [1], and deal with obstacles [2]. The study of leg dynamics is also applicable to prosthetics design and human gait
research [3].
In order to simplify the control of robotic legs and make the behaviour more tractable and intuitive, the physical model of
a leg mechanism must be simplified. Simple models, such as the inverted and double pendulum are used to approximate the
legs of walking robots[4], and the Spring Linear Inverted Pendulum (SLIP) model is used to approximate the behaviour of
running and hopping robots[2]. The controllers developed according to these simple models perform well if the actual leg
mechanism is close to the assumed model; otherwise, specific measures must be taken, depending on the differences be-
tween the models and the actual mechanisms. Examples given are the experimentally determined hip torque used in [5]
to counteract the difference between SLIP model and the actual leg, and the reduction-by-feedback strategy used in [6] to
simplify the model. Consequently, the designer is limited in the choice of mechanical solutions to those that are close to
one of the simple and well-studied models. Moreover, the dynamic models are intended for a specific motion phase, such
as stance phase, when the leg is on the ground, or swing phase, when the leg is off the ground. For robots with more than
one leg, there are more complex motion phases, each represented by a different dynamic model [7].
The presented work aims at finding simple models, using bond graphs, with approximately similar dynamic behaviour to
that of a given leg mechanism for each motion phase. These simple models are real-world mechanisms, not linearized ver-
sions of the original leg design, and provide an intuitive understanding of the behaviour of the more complex mechanism, as
well as a basis for controller design. In a previous work [8], the design of the leg mechanism is detailed, and the equations of

⇑ Corresponding author. Address: University of Toronto Institute for Aerospace Studies, 4925 Dufferin Street, Toronto, Ontario M3H 5T6, Canada. Tel.: +1
416 946 3357; fax: +1 416 946 7109.
E-mail address: [email protected] (M.R. Emami).

1569-190X/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.simpat.2013.09.006
 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143 133

motion for the system are obtained using the Lagrangian approach, but the expressions are too complex and non-intuitive to
be used for simplification studies. Instead, in here a bond graph representation of the system is derived.
Bond graphs are domain-independent graphical descriptions of dynamical behaviour of physical systems. Bond graphs
are based on the concept that energy exchange is a common notion to dynamic systems regardless of their physical do-
main[9]. Therefore, in a bond graph components are defined by their energetic behaviour; they can either supply or absorb,
store or dissipate, and reversibly or irreversibly transform energy [10]. Bond graphs can be used to represent complex three
dimensional mechanical systems [11,12], as well as mechatronic systems [13–16].
The first part of the paper, Section 2, expresses dynamics of a leg mechanism using bond graphs in a way that it facilitates
further simplifications. The second part of the paper, Section 3, applies a combination of various simplification techniques to
the bond graph model to find simpler representation of the system in each motion phase. The behaviour of simpler models is
also compared with that of the original model in Section 4. Some concluding remarks are made in Section 5.

2. Bond graph model of the leg mechanism

Bond graphs represent the system dynamics by considering the power exchange between its components. The system
variables such as force, velocity, current, and voltage, are unified into two groups, the flow and effort. By multiplying these
generalized variables, the power flow between components can be computed. The bond graph uses the dynamic equations of
its components and initial conditions to calculate the behaviour of the system [16]. Furthermore, bond graphs can be linked
together in a modular fashion to represent a more complex mechanism. For example, the rigid body in Fig. 1 is represented
using a bond graph composed of several modules. The generalized flow and effort variables correspond to velocity and force,
respectively. The velocity (flow) of point O1 is specified using flow source components (C), and the force (effort) applied at
point O2 is represented by effort source components (D). Both velocity and force vectors are expressed in the local coordinate
frame of the body. The vectors r1 and r2 define the position of points O1 and O2 with respect to the centre of mass of the body
in the local coordinate frame using transformer blocks (B). The rotational dynamics of the body are represented using an
inductor component (A), and the translational dynamics of the body are represented by two inertia components as inductors
and two effort sources in the world coordinate frame (F). A coordinate transformation block (E) relates the velocities and
forces of the body centre of mass expressed in the world and local coordinate frames.
Using the above-mentioned six blocks the planar dynamics of the rigid body can be calculated, together with the power
flow to each component. By linking a number of rigid body modules together, the dynamics of a more complex system, such
as the leg mechanism shown in Fig. 2 can be derived.
The robotic leg proposed in [8], called Linkage Leg, is shown in Fig. 2. The leg is composed of four links. The hip joint O0
connects the body to the first link, the thigh {O0O1O4}. The tibia {O1O2}, foot {O2O3O5} and tendon {O4O5}, together with
the thigh, form a four-bar linkage. The lengths a1 . . . a8 define the geometry of the four links. The two degrees of freedom
of the Linkage Leg are the angle between coordinate frames {O0} and {O1}, defined as the hip angle h1, and the angle between
the coordinate frames {O1} and {O2}, defined as the knee angle h2.

Fig. 1. Bond graph representation of a rigid body dynamics.

134 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143

Fig. 2. Linkage Leg mechanism.

A string (1), starting from point A and ending at point O5, is wrapped around pulleys at the joints O1 and O2. Point A is
chosen such that the length of the string remains constant as the knee angle h2 changes during the swing phase, so the knee
is free to contract. During the stance phase, a disk cam mechanism (4), activated by a 5 N, 12 V solenoid (3), locks the string
preventing it from rotating around the pulleys and forcing the elastic element (2) to extend and store energy. During the
swing phase the string is unlocked and the knee is free to retract.
The leg has two actuators. The hip actuator at joint O0 controls the angle of leg relative to the body of the robot. The knee
actuator at joint O1 controls the length of the leg during swing, and it is also used in parallel to the elastic element (2) to
control the height of each hop.
The Linkage Leg mechanism has a number of advantages over similar robotic legs. The leg employs only revolute joints,
making it simpler to build and potentially lighter than legs with prismatic joints [1]. The Tibia and Tendon segments are
loaded only in compression and tension, respectively, making them easier to design and lighter to build. Another significant
advantage is that the elastic element (2) needs to act at tension only, allowing the use of rubber or latex that have a higher
energy storing capacity per unit mass compared to steel springs. Further, the proposed Linkage Leg can change its knee angle
at touchdown easily by changing the timing for locking the string with no need for mechanical adjustments.
The most significant advantage of the Linkage Leg is that its dynamics can be tuned to a SLIP model for the stance phase
and to a specific double pendulum for the swing phase, using the approach that will be discussed in the sequel. This allows
for a simple and intuitive method of understanding and defining the dynamics of the leg, as well as the use of existing control
strategies for the SLIP and double pendulum [17,2].
The bond graph representation of the Linkage Leg is shown in Fig. 3. The bond graph is composed of five rigid body mod-
ules, the thigh, tibia, foot, tendon and robot body, to which the leg is attached. Additionally, a ground model is used to sim-
ulate the toe (O3) touching the ground.
The bond graph model of Linkage Leg was simulated in two different conditions: the stance phase and swing phase, and
Ò
its behaviour was compared with the models of the same mechanism developed using both MATLAB SimMechanics phys-
ical modelling as well as the m-file script based on the system differential equations as detailed in [8]. The results were iden-
tical, verifying the bond graph model.
Fig. 4 shows the trajectories of the hip and knee joints during one step. The takeoff and touchdown points are also indi-
cated. The bond graph has been simulated for a representative step motion, with the robot running at 3 m/s and reaching
1.5 m apex height.
To simulate the leg using the bond graph shown in Fig. 3, the geometric parameters in Table 3 are used, together with the
masses and moments of inertia from Table 2. The bond graph is initialized using the initial conditions in Table 1. During the
swing phase the knee is free to rotate, and during the stance phase a 38.4 N m/deg spring is applied around the knee joint.
The ground model is used during the stance phase, with a 10,000 N/m spring and 100 N/(m/s) damper holding the toe point
locked to the ground. The spring and dampers are set to zero during the swing phase so the body is free to move according to
the projectile motion.
 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143 135

Fig. 3. Bond graph representation of Linkage Leg.

3. Bond graph model simplification

3.1. Simplification algorithms

This section defines a method of determining the analogous simpler mechanisms for a given mechanism using its bond
graph model, and applies it to the Linkage Leg. The proposed method is a two-step process. The first step determines the
dynamic elements that are necessary for the behaviour to remain similar to that of the original system, and the second step
eliminates the bonds that have insignificant power flow and determines what mechanism is formed by the remaining dy-
namic elements. The term ‘‘dynamic element’’ refers to the bond graph component associated with masses and moments
of inertia, generally represented by induction components.
The first step is an element-oriented simplification approach that eliminates the maximum number of dynamic elements
while keeping the dynamic behaviour unchanged [18]. The algorithm requires a set of dynamic elements S to be simplified. If
there are m dynamic elements in the bond graph (such as masses and moments of inertia) and simplification is to be done for
n of them, then there are C mmn possible combinations. For each possible combination, the bond graph is simulated, and the
136 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143

Touchdown
1.5 Takeoff
Takeoff

vertical height (m)

1

0.5

-0.5 0 0.5 1 1.5 2 2.5

horizontal distance (m)

-40
takeoff angle
-50 touchdown angle
hip angle (deg)

hip angle

-60 Touchdown
Takeoff
Takeoff
-70

-80
Swing Stance
-90
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
time (s)

-102
-104
knee angle (deg)

-106
-108 Touchdown Takeoff
Takeoff
-110
-112
-114
-116
-118
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
time (s)

Fig. 4. Trajectories of the knee and hip joint during one step.

Table 1
Initial conditions for stance and swing phase simulations.

Vertical body velocity (m/s) Horizontal body velocity (m/s) h1(deg) h_ 1 (deg/s) h2 (deg) h2 (deg/s)

Stance phase 3 3 50 151 103 320

Swing phase 3 3 74 0 104 0

behaviour of a set of test bonds is compared, such as the bonds related to the controlled degrees of freedom of the system.
The result of this first step is a set of elements that produce the closest behaviour to the original system for each level of
simplification. The level of simplification represents the number of removed dynamic elements (level zero for no simplifica-
tion, level one for one element removed, level two for two elements removed, etc.). Beyond a certain simplification level, the
system behaviour will start markedly deviating from the original behaviour, indicating that all the remaining elements are
necessary.
The second simplification step can be applied to any of the simplification levels from the first step. The goal is to find the
necessary bonds for the system behaviour to remain relatively unchanged. First, a metric is found that represents the impor-
tance of each bond and component for the bond graph. Bonds with zero or small power flow get reduced. The metric used in
this step is based on the correlation of the energy flow pattern in the bonds of the bond graph. The end result is a rank for
 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143 137

each bond that represents the importance of that bond to the overall system behaviour. The algorithm, based on the ap-
proach presented in [19], consists of the steps shown below:

1. Find energy trajectories for each bond (Ei), and form the E matrix E = [E1,E2, . . ., En]. The algorithm can analyze the entire
bond graph or only a selection of bonds.
2. Apply Singular Value Decomposition (SVD) to the E matrix, and calculate the U, S, V matrices such that E = USV1
P !
3. Calculate vector I ¼ ni¼1 r2i j v l j, where n is the number of bonds analyzed, ri=1,. . .,n are the elements of the diagonal matrix
!
S, and v l is the ith column of vector V.
4. Normalize the vector I and order the bonds in the order of decreasing relative importance.
I
5. Choose a value r to be the cut-off threshold. Every time iþ1 Ii
< r, a new reduction stage is reached. For this analysis r = 1.5.
6. Finally, one can choose as many reduction stages as necessary. The first two or three reduction stages usually contain the
meaningful dynamics.

The rationale for the proposed two-step approach is as follows. Most methods, such as those presented in [19,20] which
form the basis for the second step of the proposed simplification approach, as well as the one in [22] are based on the notion
of eliminating the bonds with zero or small power flow. For some cases, these methods are able to find the underlying simple
mechanisms [20]. However, these methods are not suitable as a complete solution for mechanisms with parallel links, for
example the Linkage Leg with the tibia-tendon and thigh-foot pairs, as it will be discussed in sequel. Therefore, if only
the second simplification step is applied to the Linkage Leg, all the links will be considered as important and consequently
no mass will be eliminated in the simplification process. The first step of the proposed simplification approach is then for
dealing with not only the insignificant individual masses but also the parallel links of the Linkage Leg by finding the minimal
combination of masses that maintains the original system dynamics, regardless of power flow. Once the minimal set of sig-
nificant masses is found, step 2 is able to determine the bonds that are significant in transmitting the energy in the step-1
mechanism.
It should be noted that the proposed simplification approach is dependent on the system motion phase as well as the
initial conditions for each phase. In other words, the simplification should be done locally for a certain system behaviour,
and it does not cover the entire dynamics of the system, because the significance of different dynamic elements varies
in different system behaviours. Therefore, one must first define the working condition of the system with its motion
phases and initial conditions prior to performing model simplification. This will be further discussed in the following
section.

3.2. Linkage Leg model simplification

To determine the working behaviour of the Linkage Leg two simulations were initially performed for the two motion
phases. The trajectories of the hip and the knee joints were consequently obtained, and they are shown in Fig. 5, for the
swing phase, and Fig. 6, for the stance phase. Also, the proper initial conditions for both phases were obtained, as shown
in Table 1. The following model simplifications will be based on the system behaviour in stance and swing phases, which
resulted from the obtained initial conditions. This local simplification approach would provide more effective results that

-40

-104
-45
-106
-50
knee angle (deg)

-108
hip angle (deg)

-55
-110

-60
Touchdown -112

-65 Takeoff -114

-70 -116

-75 -118
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.1 0.2 0.3 0.4 0.5 0.6
time (s) time (s)

Fig. 5. Swing phase joint angle results for Case 3.

138 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143

-45 -102
Touchdown
No Simplification
-50 Simp. level 1
-104
Simp. level 2
Simp. level 3
-55 Simp. level 4
-106 Simp. level 5

knee angle (deg)

hip angle (deg)

Simp. level 6
-60
-108
-65
Takeoff
-110
-70

-112
-75

-80 -114
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
time (s) time (s)

Fig. 6. Stance phase joint angle results for Case 3.

Table 2
Variables for the test cases studied.

Inertia variable Case 1 Case 2 Case 3

m1 (kg) 0.5 1 0.5
m2 (kg) 0.3 0.5 0.5
m3 (kg) 0.2 0.5 0.5
m4 (kg) 0.1 0.1 0.2
I1 (kg m2) 0.02 0.03 0.02
I2 (kg m2) 0.015 0.02 0.02
I3 (kg m2) 0.01 0.02 0.02
I4 (kg m2) 0.007 0.007 0.02

Table 3
Geometric parameters for the Linkage Leg.

Geometric variable All cases Geometric variable All cases

a1 (m) 0.5 x1 (m) 0.25
a2 (m) 0.5 x2 (m) 0.25
a3 (m) 0.5 x3 (m) 0.15
a4 (m) 0.3 x4 (m) 0.25
a5 (m) 0.5 y1 (m) 0.00
a6 (m) 0.2 y2 (m) 0.00
/2 (deg) 0.00 y3 (m) 0.00
/3 (deg) p y4 (m) 0.00

are directly relevant to the interested working behaviour of the system. For many systems, a global approach of simplifica-
tion that can result in a unique simpler system resembling the original one at all working conditions may be inefficient or
infeasible, as it is argued in [20].
To simplify the Linkage Leg bond graph model, three cases are investigated. The test cases are meant to reflect different
construction possibilities for the Linkage Leg, based on CAD models of possible configurations. These cases, listed in Table 2,
were chosen to represent different construction methods and leg configurations. The first one represents a more natural leg,
with the thigh (m1) and tibia (m2) heavier than foot (m3) and tendon (m4). The third case represents a more classic robot
construction, with all the linkages of equal mass, with only the tendon being lighter. The second case is in between the first
and third ones, with the thigh heavier than the tibia and foot, which are equal, and the tendon lighter. In all three cases the
Linkage Leg geometry is kept constant as shown in Table 3 and the mass and moment of inertia parameters are changed, as
listed in Table 2.
 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143 139

Table 4
Element simplification results.

Simp. Level Swing phase Stance phase

m1 m2 m3 I3 m4 I4 mbody cost m1 m2 m3 I3 m4 I4 mbody cost
Case 1
Level 0 0.5 0.3 0.2 0.01 0.1 0.007 10 0 0.5 0.3 0.2 0.01 0.1 0.007 10 0
Level 1 0.5 0.3 0.2 0.01 0.007 10 0.035 0.5 0.3 0.2 0.1 0.007 10 0.002
Level 2 0.5 0.3 0.2 0.01 10 0.061 0.5 0.3 0.2 0.1 10 0.008
Level 3 0.5 0.2 0.1 10 0.123 0.5 0.3 0.2 10 0.028
Level 4 0.5 0.2 10 0.083 0.5 0.3 10 0.065
Level 5 0.3 10 0.244 0.5 10 0.113
Level 6 10 0.489 10 0.261
Case 2
Level 0 1 0.5 0.5 0.02 0.1 0.007 10 0 1 0.5 0.5 0.02 0.1 0.007 10 0
Level 1 1 0.5 0.5 0.02 0.007 10 0.022 1 0.5 0.5 0.1 0.007 10 0.003
Level 2 1 0.5 0.5 0.02 10 0.033 1 0.5 0.5 0.1 10 0.008
Level 3 1 0.5 0.5 10 0.103 1 0.5 0.5 10 0.024
Level 4 1 0.5 10 0.128 1 0.5 10 0.118
Level 5 0.5 10 0.320 1 10 0.164
Level 6 10 0.449 10 0.423
Case 3
Level 0 0.5 0.5 0.5 0.02 0.2 0.02 10 0 0.5 0.5 0.5 0.02 0.2 0.02 10 0
Level 1 0.5 0.5 0.5 0.02 0.02 10 0.119 0.5 0.5 0.5 0.2 0.02 10 0.003
Level 2 0.5 0.5 0.02 0.2 10 0.084 0.5 0.5 0.5 0.2 10 0.023
Level 3 0.5 0.5 0.02 10 0.159 0.5 0.5 0.5 10 0.055
Level 4 0.5 0.5 10 0.231 0.5 0.5 10 0.142
Level 5 0.5 10 0.518 0.5 10 0.194
Level 6 10 0.823 10 0.338

Fig. 7. Swing phase, simplification level five.

140 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143

Fig. 8. Swing phase, simplification level four.

The first step of the simplification requires a set of dynamic elements to be simplified and the bonds used to test the sim-
ilarity to the original system. Similar to most robotic legs, the Linkage Leg has two degrees of freedom, the hip (h1) and the
knee (h2) joints. These two joints are also the actuated and controlled joints. Therefore, the bonds associated with the mo-
ments of inertia of these two joints are used to test the similarity of each simplified combination to the original system. The
components associated with the other dynamic elements (masses m1, m2, m3, m4 of the Linkage Leg bodies, the mass mrobot of
the robot body and moments of inertia I3 and I4) form the set of elements to be simplified.
The results of the first simplification step, presented in Table 4, show the combination of elements with the lowest cost
(closest fit to the original data) for each simplification level. The cost function is the square difference between the trajec-
tories of joints O0 and O1 of the simplified simulation and those of the original case. These trajectories for case 3 are shown in
Fig. 5, for the swing phase, and Fig. 6, for the stance phase. The unmodified trajectory is the same as the one in Fig. 4.
The simplification levels start from level zero, with no simplified elements, and go up to level six, where all the dynamic
elements of the Linkage Leg are removed. The data shows that the mass of the robot body is the most important element in
all cases, and was never simplified.
For the swing phase, the lower simplification levels are different between the test cases, indicating that the least impor-
tant dynamic elements vary depending on the test case being considered. However, the higher simplification levels are the
same for all test cases. If four elements are simplified and two are kept intact, m1 and m3 are determined to be the important
elements, and if only one element is kept, m2 is determined to be the most influential. In Fig. 5 it can be seen that simpli-
fication levels five and six have markedly different trajectories for the joints on interest, O0 and O1, while simplification level
four shows a closer behaviour to the unmodified system. Consequently, the simplification levels four and five will be ana-
lyzed further.
It should be noted that to obtain a valid comparison of hip and joint trajectories between different simplification levels in
Fig. 5, the hip torque ship applied during the swing phase must be directly proportional with the overall moment of inertia of
the leg around the hip joint in each mechanism.
For the stance phase, Fig. 6, the Linkage Leg behaviour remains relatively unchanged no matter how many dynamic ele-
ments in the leg mechanisms are simplified. This is explained by the much larger mass assigned to the body of the robot and
the fact that the toe is in contact with the ground, which dominates the dynamic behaviour of the joints O0 and O1. The case
of all the elements being simplified is the one with the most different behaviour, as expected.
 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143 141

The next step is to understand what kind of mechanism is described by the remaining dynamic elements. For the stance
phase, the combination resulting from the sixth simplification level is investigated (all dynamic elements of the leg, but not
the robot body, are removed), whereas for the swing phase the combinations from resulting from the fourth and fifth sim-
plification levels are investigated. Given that the results are very similar for all three initial conditions cases, for the sake of
brevity they will be presented only for case 3.
The results of the second simplification step are shown in Figs. 7 and 8 for the swing phase, and, in Fig. 9 for the stance
phase. The meaningful bonds are shown in black, and the reduced bonds are greyed out. For the fifth level of simplification,
shown in Fig. 7, the gravity and inertial effects of m2 remain unchanged, together with the vector defining the location of the
centre of mass relative to O1.
The centre of mass is located on the ~ x component of the local coordinate frame, so only one of the components of the
vector is necessary. The tibia is linked to the robot body by the massless rod {O0O1}, represented in the bond graph by
the remaining elements of the body {O0O1O4}, forming a double pendulum system.
For the simplification level four, body m1 is left intact, but the bonds associated with the mass m2 are removed. Instead
the mass m3 acts at the point O2, effectively becoming part of the body {O1O2}. This also forms a double pendulum mecha-
nism, this time however both links have mass and inertia. This difference explains the marked change in behaviour between
simplification levels four and five, indicating that a double pendulum where both links have mass is the simplest mechanism
that has similar behaviour to the original Linkage Leg. The bonds associated with the position of the mass m3 relative to point
O2 have an effect on the system behaviour, moving it away from a perfect double pendulum similarity. The effect is discussed
below, in the comparison section.
The stance phase bond graph for simplification level six is shown in Fig. 9. The robot body dynamics remain unchanged,
and the other remaining elements are the ground contact model and the spring of the joint O1. The rest of the bonds are kine-
matic bonds, which describe the location of the body with respect to the ground contact point O3, as well as the orientation of
the body velocity with respect to the ground.
The resulting mechanism is similar to the SLIP model, widely used for running and hopping robots. The main difference is
that in the SLIP model the direction of the force due to the spring is coincident with the hip joint [2], whereas in the Linkage
Leg this direction is determined by the instantaneous configuration of the four bar linkage system [21], and changes depend-
ing on the leg length.

Fig. 9. Stance phase, simplification level six.

142 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143

4. Comparison between Linkage Leg and the simple mechanisms

The double pendulum is proposed as a simpler mechanism candidate for the swing phase and the SLIP model as a can-
didate for the stance phase. As indicated in the bond graph, the first rigid body of the double pendulum is composed of the
mass m1 and rotational inertia I1, and the second rigid body is composed of the mass m3 and rotational inertia I2. The com-
parison between the Linkage Leg and the double pendulum is shown in Fig. 10, and it shows that the behaviour of the hip and
knee trajectories is similar. The hip trajectory of the double pendulum has a 10% maximum difference compared with the
Linkage Leg, while the knee is closer with a 5% maximum difference.
This difference between the double pendulum and Linkage Leg is explained by the effect of the moment arm bonds on the
m3 mass. This effect can be minimized depending on the location of mass m3 relative to the point O2. A mass offset x3 of 0.5
results in the mass m3 coincident with O2, whereas an offset of 0.35 results in the mass m3 in the centre of the segment O3O5.
Both the Linkage Leg and the double pendulum trajectories end up at the same touchdown positions. The leg angles at

-35 -100

-40
-105

-45

-110
knee angle (deg)
hip angle (deg)

-50

-55 -115

-60
-120

-65
Simp. level 4
-125
-70 Equivalent double pendulum

-75 -130
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
time (s) time (s)

Fig. 10. Comparison between Linkage Leg and double pendulum.

0.3 1

0.25 0.98

0.2
0.96

0.15
Body Position X (m)

Body Position Y (m)

0.94
0.1
0.92
0.05
0.9
0
0.88
-0.05

0.86
-0.1

-0.15 0.84

-0.2 0.82
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
time (s) time (s)

Fig. 11. Comparison between Linkage Leg and SLIP model.

 V. Ragusila, M.R. Emami / Simulation Modelling Practice and Theory 40 (2014) 132–143 143

touchdown are shown to have the greatest impact on the running gait in [2], further indicating suitability of the double pen-
dulum for this particular application.
The SLIP model is composed of a massless spring between the body centre of mass and the toe contact point [2]. Given
that the SLIP model does not have the same hip and knee joints of the Linkage Leg, the two systems are compared based on
the position of the centre of mass of the robot body. The comparison is shown in Fig. 11. Both the horizontal and the vertical
position of the body are very similar between the Linkage Leg and the SLIP model, indicating that the offset between the hip
and the spring force present in the Linkage Leg has little influence on the final behaviour. The maximum difference of 0.02m
between body positions is similar to the difference found between the models shown in [19].

5. Conclusions

A method was presented that finds simpler mechanisms with a dynamic behaviour similar to that of a more complicated
robotic leg mechanism for each specific motion phase. Such simpler mechanisms will be good representatives of the original
mechanism while having the advantages of simplicity and tractability. The original leg mechanism was first modelled using a
bong graph approach, and then a two-step simplification process was used to determine simpler mechanisms with the same
significant dynamic characteristics. The first step finds the dynamic elements that are necessary for the system behaviour to
remain unchanged for each motion phase, and the second step simplifies the bond graph formed by such elements and con-
structs the resulting mechanism expressed by the simplified graph.
For the case study of a Linkage Leg two motion phases were analyzed, i.e., swing and stance. During the swing phase, two
masses elements were found to be required for generating the same behaviour as the original system, resulting in a double
pendulum mechanism. And, for the stance phase the resulting simplified mechanism was found to be the classic SLIP model
that has been used for modelling robotic and biologic leg dynamics.
Having such simpler mechanisms will enable the designer to fine-tune the kinematic, dynamic and control parameters of
the original mechanism in a more intuitive way using methods such as the one discussed in [8].

References

[1] B. Brown, G. Zeqlin, The bow leg hopping robot, in: IEEE International Conference on Robotics and Automation, Leuven, Belgium, 1998, pp. 781–786.
[2] M.H. Raibert, Legged robots, Communications of the ACM (1986) 499–514.
[3] Manoj Srinivasan, Fifteen observations on the structure of energy-minimizing gaits in many simple biped models, Journal of the Royal Society Interface
8 (54) (2011) 74–98.
[4] Steve Collins, Andy Ruina, Russ Tedrake, Martijn Wisse, Efficient bipedal robots based on passive dynamic walkers, Science Magazine (2005) 1082–
1085.
[5] Sang-Ho Hyon, Development of a Biologically Inspired Robot Kenken, in: Proceedings 2002 IEEE International Conference on Robotics and Automation,
Washington DC, USA, 2002, pp. 3984–3991.
[6] I. Poulakakis, Stabilizing Monopedal Robot Running: Reduction-by-feedback and Compliant Hybrid Zero Dynamics, The University of Michigan, PhD
Thesis, 2008.
[7] Andrew Biewener, Patterns of mechanical energy change in tetrapod: pendula, springs and work, Journal of Experimental Zoology Part A: Comparative
Experimental Biology (2006) 899–911.
[8] Victor Ragusila, Reza R. Emami, A novel robotic leg design with hybrid dynamics, International Journal of Humanoid Robotics (2012).
[9] H.M. Paynter, Analysis and Design of Engineering Systems, M.I.T. Press, Cambridge, MA, USA, 1961.
[10] W. Borutzky, Bond graph modelling and simulation of mechatronic systems: an introduction into the methodology, in: 20th European Conference on
Modelling and Simulation ECMS 2006. Modelling Methodologies and Simulation Key Technologies in Academia and Industry, Sankt Augustin, 2006, pp.
17–28.
[11] M.J.L. Tiernego, A.M. Bos, Modeling the dynamics and kinematics of mechanical systems with multibond graphs, Journal of the Franklin Institute 319
(1-2) (1985) 37–50.
[12] Jinhee Jang, Changsoo Han, Proposition of a modeling method for constrained mechanical systems based on the vector bond graph, Journal of the
Franklin Institute 335B (3) (1998) 451–469.
[13] P.J. Gawthrop, Bond graphs: a representation for mechatronic systems, Mechatronics 1 (2) (1991) 127–156.
[14] R. Fotsu-Ngwompo, S. Scavarda, D. Thomasset, Bond graph methodology for the design of an actuating system: application to a two-link manipulator,
in: 1997 IEEE International Conference on Systems, Man, and Cybernetics. Computational Cybernetics and Simulation, Orlando, 1997, pp. 2478–2483.
[15] J. van Amerongen, Mechatronic design, Mechatronics 13 (10) (2003) 1045–1066.
[16] Dean C. Karnopp, Donald L. Margolis, C. Rosenberg, System Dynamics: Modeling and Simulation of Mechatronic Systems, 4th ed., John Wiley & Sons,
Hoboken, United States of America, 2006.
[17] Jerry Pratt, Exploiting Inherent Robustness and Natural Dynamics in the Control of Bipedal Walking Robots, Ph.D. dissertation, Comp. Sci. Dept.
Massachusetts Institute of Technology, Cambridge, Massachusetts, 2000.
[18] Y.H. Hung, C.H. Wu, C.W. Hong, An element-oriented model simplification algorithm based on dynamic similarity, in: Proceedings of the Institution of
Mechanical Engineers. Part I: Journal of Systems and Control Engineering, 2012, pp. 56–69.
[19] Tulga Ersal, Fathy K. Hosam, Jeffrey L. Stein, Realization-preserving structure and order reduction of nonlinear energetic system models using energy
trajectory correlations, Journal of Dyanmic Systems, Measurement and Control 131 (3) (2009) 031004. 8 pp..
[20] T. Ersal, H.K. Fathy, J.L. Stein, Structural simplification of modular bond-graph models based on junction inactivity, Simulation Modelling Practice and
Theory 17 (1) (2009) 175–196.
[21] W.L. Cleghorn, Mechanics of Machines, first ed., Oxford University Press, 2005.
[22] D.G. Rideout, J.L. Stein, L.S. Louca, Systematic identification of decoupling in dynamic system models, Transactions of the ASME. Journal of Dynamic
Systems, Measurement and Control (2007) 503–513.