Robotics and Autonomous Systems 116 (2019) 1–23

Contents lists available at ScienceDirect

Robotics and Autonomous Systems

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

Real-time walking pattern generation for a lower limb exoskeleton,

implemented on the Exoped robot
∗
Jafar Kazemi, Sadjaad Ozgoli
Department of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran

highlights

• A real-time walking pattern generation method is provided.

• The trajectory planning is formulated as a feedback control problem.
• A solution is provided for an optimal control problem with changeable final states.
• The backward balance of the lower limb exoskeleton is considered in the walking pattern.

article info a b s t r a c t

Article history: Lower extremity exoskeletons have been developed as a motion assistive technology in recent years.
Received 30 September 2018 Walking pattern generation is a fundamental topic in the design of these robots. The usual approach
Received in revised form 10 January 2019 with most exoskeletons is to use a pre-recorded pattern as a look-up table. There are some deficiencies
Accepted 18 February 2019
with this method, including data storage limitation and poor regulation relating to the walking
Available online 2 March 2019
parameters. In addition, the walking parameters can be taken in hand very hard. Therefore modeling
Keywords: the human walking pattern is required. The few existing models provide piece by piece walking
Exoskeleton patterns, only generating at the beginning of each stride. In this paper, a real-time walking pattern
Walking pattern generation method is provided which enables changing the parameters during the stride. For this
Optimal control purpose, two feedback controlled third order systems are proposed as real-time trajectory planners for
Center of mass
generating the trajectories of the x and y components of each joint’s position. The boundary conditions
Exoped
of the trajectories are determined to prevent backward balance loss by appropriate displacement of the
center of mass. In addition, a cost function is intended for each trajectory planner in order to increase
the smoothness of trajectories. Optimization technique is used to design the feedback controller for
tracking the boundary conditions in such a way that the cost function is minimized. Finally, the proper
joint angles are generated using inverse kinematics transformation. The performance of the proposed
pattern generator is verified via real experiments on the Exoped robot.
© 2019 Elsevier B.V. All rights reserved.

1. Introduction walking patterns, depending on the application and structure

of the exoskeleton. These methods can be classified into three
Increase in diseases and accidents related to human mobility, categories: model-based, sensitivity amplification, and predefined
in addition to population aging, has caused much attention to gait trajectory strategies [4]. The model-based strategies take
be given to motion assistive technologies including wearable stability into account in order to determine a walking pattern.
robots; particularly lower limb exoskeletons. Research on pow- The zero moment point (ZMP) is the most widely used stability
ered human exoskeleton devices dates back to the 1960s in the criterion in gait planning [5,6]. This method relies on the accuracy
United States [1] and in the former Yugoslavia [2] for military of the human-exoskeleton model and requires various sensors.
and medical service purposes respectively. Since then, exoskele- Sensitivity amplification strategies are applied to the exoskele-
ton robots have been well-developed, particularly for medical tons in which user applies force to the robot. These include
purposes. Some of them have even hit the market [3]. load carrying [7] and rehabilitation [8] robots. Finally, predefined
One of the major challenges in designing these robots is walk- gait trajectory strategies use a pre-recorded pattern of a healthy
ing pattern generation. There are various methods for planning person as the reference trajectory. These are proper mostly for
robots that a patient is the user.
∗ Corresponding author. Due to the difficulty of accurate ZMP measurement, model-
E-mail address: [email protected] (S. Ozgoli). based strategies are not cost-efficient. Furthermore, sensitivity

https://doi.org/10.1016/j.robot.2019.02.012
0921-8890/© 2019 Elsevier B.V. All rights reserved.
2 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

amplification methods are not suitable for the kind of assis-

tance which paraplegic patients need. Therefore, predefined gait
trajectory methods are the usual approach in most of medical
exoskeletons such as ReWalk [9], eLEGS [10], and ATLAS [11],
which is aimed at subjects who are losing their ability to move.
There are some shortcomings with this method such as data
storage limitation and pattern adjustment depending on different
walking parameters and different individuals. Motivated by the
above problems, a number of simple models have been devel-
oped to generate walking patterns alternated to pre-recorded
gaits. The common procedure for these models is to determine
the boundary conditions of the joints’ trajectory at some speci-
fied via-points and to fit a mathematical curve (e.g. polynomial
or sinusoidal) to them. In [12] and [13], the hip trajectory is
approximated by polynomial and sinusoidal function segments
respectively. In [14], two sinusoidal trajectories are proposed as
the x and y components of ankle position. In [15] polynomial and
sinusoidal functions are used for generating the joint positioning
of a biped. Also, spline interpolation are used in [16] for joint
motion planning of a wearable robot by satisfying determined
via-point constraints.
In previous works, the trajectories are generated piece by
piece as the segments between via-points and the endpoint
boundary conditions are fixed within the segments. In other
words, the walking patterns are determined at the via-point times
and changing walking parameters during a stride is not possible.
While, for real-time control of the robots, the walking parameters
may be updated at any time in facing obstacles or difficult
situations. Therefore, an online pattern generator is required to
adjust the trajectories according to the changes in the parameters.
In this paper, we propose a real-time minimum jerk walking
pattern generator for lower limb exoskeletons. In the proposed
method, the walking parameters (including step length, maxi-
mum foot clearance, and stride time) can be changed during
the stride without discontinuity in the second derivative of the Fig. 1. The Exoped robot and pilot.
trajectories. The idea of our method is to consider the trajectory
as the output of a third order system. A feedback controller is then
designed to impose the system’s output to satisfy the boundary 2. The Exoped robot
conditions of the trajectory. By introducing a minimum jerk cost
function, the trajectory planning problem is formulated as an Exoped is a 4 DOF lower limb exoskeleton. It has 2 DOFs on
optimal control problem with changeable final states. A solution each leg: 1 at the hip and 1 at the knee, driven by 4 brushless
for this problem is proposed using the minimum principle of electronically communicated (EC) motors (Fig. 1). The ankle joints
Pontryagin [17]. behave passively with a spring and their default angles are set
It should be noted that the balance and the safety of the user at 90◦ relative to the leg. The motors from the Maxon company
is the most important concern in design of the medical lower (model ‘‘EC 90 flat’’), fed by 36 V power supply, are used. Each
limb exoskeletons. Stability of the robot is guaranteed when motor is coupled with a 1:135 gearbox and internal hall sen-
sors are used to indicate the position. The control algorithm of
using the exoskeleton as a walking training assistance between
the robot including state machine, walking pattern generation,
parallel bars. While, for the personal applications and outside of
and PID controller of the motors is realized in Simulink. The
the laboratories where the wearer is supported by crutches or designed Simulink model is then converted to C code using the
walkers, the balance is maintained only in the forward direction. Waijung library block set [18] and implemented on Stm32f429
Therefore, there is not a support tool in the backward direction microcontroller.
and the backward balance should be considered in the gait. In The forward kinematics of Exoped can be described as follows:
order to ensure the dynamic stability of the robot, we set the
xR = lT sin (θRh ) + lS sin (θRh + θRk )
boundary conditions of the hip position in such a way to prevent
a backward balance loss. yR = −lT cos (θRh ) − lS cos (θRh + θRk )
The remainder of this paper is organized as follows: Section 2 xL = lT sin (θLh ) + lS sin (θLh + θLk )
introduces the Exoped robot as our implementation platform. In yL = −lT cos (θLh ) − lS cos (θLh + θLk ) (1)
Section 3, the real-time trajectory planners are designed for each
x and y component of the position of the joints. In Section 4, the where R and L refer to right and left leg, respectively. θ∗h and θ∗k
boundary conditions of the trajectories are determined in such denote the angles of the hip and knee joint of each leg. lT and lS
a way to avoid a backward balance loss. The simulations and represent the length of thigh and shin respectively and xR , yR , xL ,
and yL are expressed by:
experimental results are presented in Section 5. Finally, the paper
is concluded in Section 6. xR = xRa − xh
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 3

The pattern generator consists of independent trajectory plan-

ners that generate the trajectory of the x and y components
of each joint’s position (Fig. 4). A boundary condition generator
calculates the endpoint conditions of the trajectories in order to
satisfy the walking constraints and maintain the balance of the
robot. The vectors Xj (t) and Yj (t) are defined as
Xj (t) = [xj (t) ẋj (t) ẍj (t)]T
Yj (t) = [yj (t) ẏj (t) ÿj (t)]T (4)
where the symbol j indicates the joints and j ∈ {h, Ra, La}. As
shown in Fig. 4 the boundary conditions are sufficient for plan-
ning the trajectory of xj (t). While, for generating the trajectory of
yj (t), the peak value of the trajectory is also required to define
the height of the joint in addition to the boundary conditions.
p
The peak value is denoted by yj which equals to Hs for the ankle
joints and is determined for the hip joint according to the walking
parameters.
Continuity, smoothness, and taking walking constraints into
account are the main objectives to be considered in trajectory
planning. In most previous works on pattern generation, the
walking parameters are considered to be constant during the
stride. The most common method of trajectory planning is fitting
a mathematical curve (e.g. polynomial or sinusoidal) to some
boundary conditions obtained by walking parameters. From a
systems point of view, this kind of trajectory planner can be
described as an input–output zero order system (Fig. 5). With
Fig. 2. Robot parameters description.
this, discontinuity of the inputs caused by any change in walking
parameters yields to discontinuity in the output.
By using the control scheme depicted in Fig. 3, the walking
yR = yRa − yh parameters may be updated at any time due to a change in
xL = xLa − xh the stability of the robot or the particular conditions. Therefore,
the above-mentioned trajectory planner (as in Fig. 5) cannot be
yL = yLa − yh (2)
used and an online trajectory planner is required to adjust the
−
→ −
→ −
→
in which rh = (xh , yh ) , rRa = (xRa , yRa ), and rLa = (xLa , yLa ) trajectory according to the change in the parameters in order to
represent the position of hip, right ankle and left ankle in sagittal maintain the second derivative continuity. For this purpose, we
plan, respectively. The initial position of the right ankle is defined propose feedback controlled third order systems as the online
trajectory planners for generating the trajectory of the joints’ x
as the origin point of coordination. The defined parameters are
and y component, as follows.
shown in Fig. 2.
The inverse kinematics equations of the robot are described as
3.1. The x component of the joints
the following:
lT 2 + lS 2 − xL 2 − yL 2 Fig. 6 Shows a general trajectory shape for the x component of
( )
θLk = cos−1 − 180 the position of the joints, starting from the initial condition Xj (t0 )
2lT lS
( ) converging to the final value Xj (tf ), with the minimum curvature.
xL 2 + yL 2 + lT 2 − lS 2
( )
− xL In order to generate this type of trajectory, we propose the
θLh = tan −1
+ cos −1
√ feedback controlled third order system depicted in Fig. 7, which
yL 2lT xL 2 + y L 2
can be formulated as follows:
2 2 2 2
( )
lT + lS − xR − yR ⎧
θRk = cos−1 − 180 ⎪ ẋ1 = x2
2lT lS ⎪
⎨
ẋ2 = x3
( ) Sx : (5)
xR 2 + yR 2 + lT 2 − lS 2
( )
− xR ⎪ẋ3 = u
θRh = tan −1
+ cos −1
√ . (3) ⎪
⎩
yR 2lT xR 2 + yR 2 xj = x1
It is obvious that a finite u yields a continuous trajectory with
3. Real-time trajectory planning a continuous second derivative. In addition, the cost function
denoted by Jx is intended to be minimized in order to increase
Fig. 3 shows the overall control schematic of the robot. As the smoothness of the trajectory.
shown, a well-defined algorithm calculates the walking param- tf
)2
d3 xj
∫ (
eters according to the stability of the robot and particular condi- Jx = dt (6)
tions; e.g. patient’s dimension and environment conditions. The t0 dt 3
walking parameters are used as the inputs for a pattern generator As a result, the real-time trajectory planner can be designed by
block. These parameters are step length, maximum foot clearance, calculating the feedback control law u to move the states of the
and stride time denoted by Ls , Hs , and Ts , respectively. The pattern system Sx to the changeable final condition Xj (tf ) in such a way
generator provides the appropriate position of the joints in the that the cost function Jx is minimized. By ‘‘changeable’’ we mean
sagittal plane and subsequently, the positions are transformed it can be changed during the time. We have presented Theorem 1
into joint angles using the inverse kinematics equations. Finally, a to solve this optimal control problem. Lemma 1 is used to prove
PID feedback controller is employed to regulate the joint angles. Theorem 1.
4 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 3. The overall control schematic of the exoskeleton robot.

Fig. 4. Pattern generator block diagram.

( )
12 2x3 (tf ) − 3x3 (t0 )
+
(tf − t0 )2
60(x1 (tf ) − x1 (t0 )) 12(2x2 (tf ) + 3x2 (t0 ))
c= −
(tf − t0 )3 (tf − t0 )2
3(x3 (tf ) − 3x3 (t0 ))
+ . (8)
Fig. 5. The most common trajectory planner. Xj (t0 ) and Xj (tf ) refer to the (tf − t0 )
boundary conditions of the trajectory.

Proof. According to the minimum principle of Pontryagin [17],

minimization of Jx can be achieved by minimizing the Hamiltonian
Lemma 1. u(t) = vx (t) steers the states of the system Sx from function defined as
the initial value Xj (t0 ) = [x1 (t0 ) x2 (t0 ) x3 (t0 )]T to the final value
H Xj (t ) , vx (t ) , P (t ) =vx2 (t ) + p1 (t ) x2 (t )
( )
Xj (tf ) = [x1 (tf ) x2 (tf ) x3 (tf )]T in such a way that the cost function
Jx has been minimized. vx (t) is obtained as the following: + p2 (t ) x3 (t ) + p3 (t ) vx (t ) (9)
T
vx (t ) = a (t − t0 ) − b (t − t0 ) + c
2
(7) where P(t) = [p1 (t) p2 (t) p3 (t)] is defined as co-state vector.
The optimal trajectories Xj∗ (t ) and vx∗ (t ) can be achieved by
where satisfying the following conditions:
360(x1 (tf ) − x1 (t0 )) 180(x2 (tf ) + x2 (t0 )) ∂H ( ∗
Ẋj∗ (t ) = Xj (t ) , vx∗ (t ) , P ∗ (t ) , t
)
a= −
(tf − t0 )5 (tf − t0 )4 ∂P
30(x3 (tf ) − x3 (t0 )) ∂H ( ∗
Ṗ (t ) = −
∗
X (t ) , vx∗ (t ) , P ∗ (t ) , t
)
+ ∂ Xj j
(tf − t0 )3
360(x1 (tf ) − x1 (t0 )) 24(7x2 (tf ) + 8x2 (t0 )) ∂H ( ∗
Xj (t ) , vx∗ (t ) , P ∗ (t ) , t
)
b= − 0= (10)
(tf − t0 )4 (tf − t0 )3 ∂vx
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 5

Fig. 6. A general trajectory shape for the x component of the position of the joints.

Fig. 7. The proposed trajectory planner for the x component of the position of the joints.

where the symbol * refers to the extremals of Xj (t ), vx (t ), and Therefore, from Lemma 1 and (13), ux (t) is obtained as (12).
P (t ). The necessary conditions for optimality can be written as The trajectory obtained by applying ux (t) to the system Sx is
ẋ∗1 (t) = x∗2 (t) denoted by Tx (t , tf , Xj (tf )), where:

ẋ∗2 (t) = x∗3 (t) Tx (t , tf , Xj (tf )) = xj (t)|u = ux (t ) , t < tf .

{ }
(14)
ẋ∗3 (t ) = −0.5p∗3 (t)
ṗ∗1 (t) = 0 3.2. The y component of the joints
ṗ∗2 (t ) = −p∗1 (t )
ṗ∗3 (t ) = −p∗2 (t ) Fig. 8 Shows a general trajectory shape for the y component
of the position of the joints, rising from the initial condition
vx (t ) = −0.5p3 (t ) .
∗ ∗
(11) p
Yj (t0 ) = [yj (t0 ) ẏj (t0 ) ÿj (t0 )]T to a peak of yj , and converging
By applying the initial condition Xj (t0 )[ and ]the final value T
to Yj (tf ) = [yj (tf ) ẏj (tf ) ÿj (tf )] , with the minimum curvature.
Xj (tf ), the optimal control function for t ∈ t0 , tf is obtained as We propose the feedback controlled third order system de-
(7). picted in Fig. 9 in order to plan this trajectory, which can be
Lemma 1 represents an open-loop control method to steer formulated as follows:
the states of the system Sx from a specific initial condition to a ⎧
fixed final value along with minimizing Jx[. In this ẏ1 = y2
] control method, ⎪
vx (t) is determined in t = t0 for t ∈ t0 , tf interval and the
⎪
ẏ2 = y3
⎨
final value is not changeable in t ∈ (t0 , tf ]. Using a closed-loop Sy : (15)
ẏ = u
⎩ 3
⎪
control method and online calculation of u makes the final value
⎪
yj = y1
changeable. For this purpose, Theorem 1 is represented.
In order to increase the trajectory’s smoothness, the cost func-
Theorem 1. Feedback control law u(t) = ux (t) steers the states tion denoted by Jy is intended as (16). Minimizing Jy causes a
of the system Sx from any initial value to the final value Xj (tf ) = peak on the trajectory in addition to the smoothness increment,
[x1 (tf ) x2 (tf ) x3 (tf )]T = [xj (tf ) ẋj (tf ) ẍj (tf )]T , in such a way wherein the parameter k determines the value of the peak.
that the cost function Jx is minimized. ux (t) is obtained for t < tf
tf
)2
as the following: d3 yj
∫ (
Jy = + kyj (t) dt . (16)
60(xj (tf ) − x1 (t)) 12(2ẋj (tf ) + 3x2 (t)) dt 3
ux ( t ) = − t0
(tf − t)3 (tf − t)2 We have presented Theorem 2 to design the proper feedback
3(ẍf − 3x3 (t)) control law u in order to move the states of the system Sy to the
+ . (12)
(tf − t) final condition Yj (tf ), along with minimizing the cost function Jy .
Lemma 2 is used to prove of Theorem 2.
Proof. Consider the state feedback Xj (t) = [x1 (t) x2 (t) x3 (t)]T
as the initial condition of the system Sx at any moment. The online Lemma 2. u(t) = vy (t) steers the states of the system Sy from
calculation of u in respect to the new defined initial condition the initial value Yj (t0 ) = [y1 (t0 ) y2 (t0 ) y3 (t0 )]T to the final
allows the final value Xj (tf ) to be changeable. Considering Xj (t) as value Yj (tf ) = [y1 (tf ) y2 (tf ) y3 (tf )]T in such a way that the cost
Xj (t0 ) corresponds to putting t instead of t0 in the formulation of function Jy is minimized. vy (t) is obtained as the following:
vx (t) in Lemma 1. In other words:
k
ux (t ) = {vx (t)|t0 = t } (13) vy (t ) = (t − t0 )3 + a (t − t0 )2 − b (t − t0 ) + c (17)
12
6 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 8. The general trajectory shape for the y component of the position of the joints.

Fig. 9. The proposed trajectory planner for the y component of the position of the joints.

where that the cost function Jy is minimized. uy (t) is obtained for t < tf as
360(y1 (tf ) − y1 (t0 )) 180(y2 (tf ) + y2 (t0 )) the following:
a= −
(tf − t0 )5 (tf − t0 )4 60(yj (tf ) − y1 (t)) 12(2ẏj (tf ) + 3y2 (t))
uy (t ) = −
30(y3 (tf ) − y3 (t0 )) k(tf − t0 ) (tf − t)3 (tf − t)2
+ −
(tf − t0 )3 8 3(ÿj (tf ) − 3y3 (t)) K (tf − t)3
+ − . (21)
360(y1 (tf ) − y1 (t0 )) 24(7y2 (tf ) + 8y2 (t0 )) (tf − t) 240
b= −
(tf − t0 )4 (tf − t0 )3
Proof. As in proof of Theorem 1:
12(2y3 (tf ) − 3y3 (t0 )) k(tf − t0 )2
+ −
uy (t ) = vy (t)|t0 = t
{ }
(tf − t0 )2 20 (22)
60(y1 (tf ) − y1 (t0 )) 12(2y2 (tf ) + 3y2 (t0 )) From Lemma 2 and (22), uy (t) is obtained as (21).
c= −
(tf − t0 )3 (tf − t0 )2
Calculation of k:
3(y3 (tf ) − 3y3 (t0 )) k(tf − t0 )3 According to the designed feedback control law (21) we have
+ − . (18)
(tf − t0 ) 240 tf tf
(tf − t0 )7
∫ ⏐ ∫ ⏐
yj (t ) dt ⏐⏐ yj (t ) dt ⏐⏐
⏐ ⏐
− = −k∗ × (23)
t0 k=k∗ t0 k=0
5 × 8!
Proof. As in proof of Lemma 1, the Hamiltonian function is
defined as where k∗ denotes the proper value of k for generating a trajectory
p
with a peak of yj .
H Yj (t ) , vy (t ) , P (t ) =vy2 (t ) + ky1 (t ) + p1 (t ) y2 (t )
( )
By describing the trajectories by piecewise fifth-order polyno-
+ p2 (t ) y3 (t ) + p3 (t ) vy (t ) (19) mials, the approximate integral of yj (t) for k = 0 and for k = k∗
can be calculated numerically as
The necessary conditions for optimality can be written as ⎡ ⎤T ⎡ ⎤
∗ ∗ 60yj (t0 ) (tf − t0 )
ẏ1 (t) = y2 (t) ⎢ ⎥ ⎢ ⎥
⎢ 12ẏj (t0 ) ⎥ ⎢(tf − t0 )2 ⎥
ẏ∗2 (t) = y∗3 (t) ∫ tf
⏐ ⎢ ⎥ ⎢ ⎥
1 ⎢ ÿj (t0 ) ⎥ ⎢(tf − t0 )3 ⎥
⎢ ⎥ ⎢ ⎥
ẏ∗3 (t ) = −0.5p∗3 (t) yj (t ) dt ⏐⏐ ∼ ⎥ .⎢
⏐
= ⎢ ⎥ (24)
t0 120 ⎢ 60yj (tf ) ⎥ ⎢ (tf − t0 ) ⎥
ṗ∗1 (t ) = −k k=0 ⎢ ⎥ ⎢ ⎥
⎣−12ẏj (tf )⎦ ⎣(tf − t0 )2 ⎦
⎢ ⎥ ⎢ ⎥
ṗ∗2 (t ) = −p∗1 (t )
ṗ∗3 (t ) = −p∗2 (t ) ÿj (tf ) (tf − t0 )3

vy∗ (t ) = −0.5p∗3 (t ) . (20) ⎡ ⎤T ⎡ ⎤

60yj (t0 ) (tp − t0 )
By applying the initial condition Yj (t0 ) and the final value
[ Yj](tf ), ⎢ ⎥ ⎢ ⎥
⎢12ẏj (t0 )⎥ ⎢(tp − t0 )2 ⎥
the optimal control function can be obtained for t ∈ t0 , tf as ⎢ ⎥ ⎢ ⎥
tf
∫ ⏐
(17). 1 ⎢ ÿj (t0 ) ⎥ ⎢(tp − t0 )3 ⎥
⎢ ⎥ ⎢ ⎥
yj (t ) dt ⏐⏐ ∼ .
⏐
=
120 ⎢ 60yp ⎥ ⎢ (tp − t0 ) ⎥
⎢ ⎥ ⎢ ⎥
t0 k=k∗
Theorem 2. Feedback control law u (t ) = uy (t) steers the states ⎢ j ⎥ ⎢ ⎥
⎣ 0 ⎦ ⎣(tp − t0 )2 ⎦
⎢ ⎥ ⎢ ⎥
of the system Sy from any initial value to the final value Yj (tf ) =
[y1 (tf ) y2 (tf ) y3 (tf )]T = [yj (tf ) ẏj (tf ) ÿj (tf )]T , in such a way 0 (tp − t0 )3
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 7
⎡ ⎤T ⎡ ⎤
p
60yj (tf − tp )
⎢ ⎥ ⎢ ⎥
2⎥
0 ⎥ ⎢ f − tp ) ⎥
(t
⎢ ⎥ ⎢
⎢
1 ⎢ 0 ⎥ ⎢(tf − tp )3 ⎥
⎢ ⎥ ⎢ ⎥
+ ⎢ ⎥ .⎢ ⎥ (25)
120 ⎢ 60y (t ) ⎥ ⎢ (tf − tp ) ⎥
⎢ j f ⎥ ⎢ ⎥
⎣−12ẏj (tf )⎦ ⎣(tf − tp )2 ⎦
⎢ ⎥ ⎢ ⎥

ÿj (tf ) (tf − tp )3

where the parameter tp refers to the peak time and is obtained
approximately by
p
yj − yj (t0 )
tp ∼
= t0 + p (tf − t0 ) (26)
2yj − yj (t0 ) − yj (tf )

Using (23), (24), and (25) the value of k∗ can be determined.

The trajectory
( obtained by applying ) uy (t) to the system Sy is
p
denoted by Ty t , t0 , tf , Yj (t0 ), Yj (tf ), yj , where:
p
Ty t , t0 , tf , Yj (t0 ), Yj (tf ), yj = yj (t)|u = uy (t ) , t < tf .
( ) { }
(27)

Tx and Ty are employed as real-time smooth trajectory plan-

ners for planning the trajectory of the x and y component of the Fig. 10. Parameters description in a stride cycle . (For interpretation of the
position of each joint respectively. The only parameters needed references to color in this figure legend, the reader is referred to the web version
of this article.)
for planning the trajectories are (tf , Xj (tf )) for Tx , and (t0 , tf , Yj (t0 ),
p
Yj (tf ), yj ) for Ty , which can be extracted as the endpoint boundary
conditions of the trajectories.
In Section 4, the endpoint boundary conditions are calculated where l = lS + lT . The parameters rL (t ) and rR (t ) are defined as
for each joint in respect to walking constraints and to prevent rL (t ) = (xh (t) − xLa (t))2 + (yh (t) − yLa (t))2
a backward balance loss. For this purpose, the gait cycle is di-
vided into two phases: the double support phase and the single rR (t ) = (xh (t) − xRa (t))2 + (yh (t) − yRa (t))2 . (29)
support phase. The boundary conditions of the double support
phase are determined to avoid a backward balance loss; and In order to generating walking pattern, the boundary condi-
the endpoint boundary conditions are calculated for the single tions of the joints should be determined for each phase (double
support phase in order to satisfy the walking parameters and support phase: t0 ≤ t < t0 + 0.2Ts , single support phase:
constraints. t0 + 0.2Ts ≤ t < t0 + Ts ).

4. Endpoint boundary conditions 4.1. Double support phase

At the end of the double support phase, by ignoring the effect

In medical lower limb exoskeleton, the patient uses parallel
of the crutches, the robot can be regarded as an inverted pendu-
bars, walkers or crutches as he/she walks. When using parallel
lum at t = td as shown in Fig. 11. For the sake of simplicity, the
bars, the stability is guaranteed in both forward and backward
effect of patient interaction with robot is considered in a simple
direction. But for the case of using walkers or crutches, the walk-
way. In fact, as the patient is paraplegic and his/her legs do not
ing pattern should be able to avoid a backward balance loss. The
apply torque, it is a good approximation to model it just as mass
well-known ZMP stability criterion cannot be used confidently
and moment of inertia. Assuming that the CoM is approximately
for the motion planning because the disturbance coming from
located on the hip, the following constraint should be satisfied
the user has a significant effect on the zero moment point. By
to avoid a backward balance loss at the beginning of the single
modeling the robot as an inverted pendulum, the body center of
support phase [16]:
mass (CoM) is an alternative index for assessing the stability. To
ẋh (td ) > ω(xLa0 − xh (td )), ω= /
√
prevent falling backward at the beginning of the single support g l (30)
phase, the velocity of the CoM must be greater than a critical
where ω is the natural frequency of the pendulum. On the other
value. For this purpose, a double support phase (approximately
hand, it can be estimated that
20% of the gait cycle) is required in order to increase the velocity
from zero to the critical value. Fig. 10 demonstrates the funda- xh (td ) − xh (t0 ) ≈ 0.1Ts × ẋh (td ). (31)
mental parameters involved in a stride cycle beginning from
Using (30) and (31), the position and the minimum velocity of
the initial ground contact position denoted by xRa0 and xLa0 . In
the hip in the horizontal direction at t = td can be determined as
this stride, the right leg (Red) swings while the left one (Black)
stays on the ground. The robot’s posture is displayed at different xRa0 + xLa0 (1 + 0.2ωTs )
xh (td ) = (32)
moments demonstrating the walking constraints, which are given 2 + 0.2ωTs
as
ẋh (td ) = ω(xLa0 − xh (td )). (33)
rL (t) = l for t = t0 , t0 + Ts
2
Regarding to Fig. 11, by using Euler’s equation, the tangential
rR (t) = l2 for t = t0 , td , t0 + Ts acceleration of the center of mass at t = td can be obtained as
max(yh ) = l g(xLa0 − xh (td ))
aT (td ) = √ . (34)
max (yRa ) = Hs (28) (xLa0 − xh (td ))2 + y2h (td )
8 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Table 1
Boundary conditions of the x component of the joints position for the double
support phase (tf = t0 + 0.2Ts ).
Joint Hip Left ankle Right ankle
xRa0 +xLa0 (1+0.2ωTs )
xj (tf ) 2+0.2ωTs
xLa0 xRa0
ω xLa0 − xh (tf )
( )
ẋj (tf ) 0 0
g (x −x (t ))×y (t )
ẍj (tf ) − La0 h f 2 2h f 0 0
(xLa0 −xh (tf )) +yh (tf )

Table 2
Boundary conditions of the y component of each joints position for the double
support phase (tf = t0 + 0.2Ts ).
Joint Hip Left ankle Right ankle
√ ( )2
yj (tf ) l2 − xh (tf ) − xRa0 0 0
xRa0 −xh (tf )
ẏj (tf ) yh (tf )
× ẋh (tf ) 0 0

( )
g xLa0 −xh (tf )
2
ÿj (tf ) − 0 0
(xLa0 −xh (tf ))2 +y2h (tf )
p
√
yj l2 − 0.25 (xLa0 − xRa0 )2 0 0

Table 3
Boundary conditions of the x component of the joints position for the single
support phase (tf = t0 + Ts ).
Joint Hip Left ankle Right ankle
xj (tf ) xLa0 + 0.25Ls xLa0 xLa0 + 0.5Ls
ẋj (tf ) 0 0 0
ẍj (tf ) 0 0 0

constraints (28):

xh (t0 + Ts ) = xLa0 + 0.25Ls (41)

Fig. 11. The inverted pendulum model of the robot at t = td .
√
yh (t0 + Ts ) = l2 − (0.25Ls )2 . (42)
In addition, the following constraints should be imposed to
By setting the radial component of the center of mass accel- prevent an ill-posed inverse kinematics problem caused by an
eration to zero at t = td , the value of ẍh (td ) and ÿh (td ) can be excessive distance between the hip and the ankle joints
defined as
ṙL (t0 + Ts ) = 0 (43)
g(xLa0 − xh (td )) × yh (td )
ẍh (td ) = − (35) ṙR (t0 + Ts ) = 0 (44)
(xLa0 − xh (td ))2 + y2h (td )
g(xLa0 − xh (td )) 2 r̈L (t0 + Ts ) < 0 (45)
ÿh (td ) = − . (36)
(xLa0 − xh (td ))2 + y2h (td ) r̈R (t0 + Ts ) < 0. (46)
The following constraints should be imposed to prevent an Given that ẋLa (t0 + Ts ) = ẋRa (t0 + Ts ) = ẏLa (t0 + Ts ) =
ill-posed inverse kinematics problem caused by an excessive dis- ẏRa (t0 + Ts ) = 0, and using (43) and (44), it can be shown that
tance between the hip and the right ankle.
ẋh (t0 + Ts ) = 0 (47)
rR (td ) = l2 (37)
ẏh (t0 + Ts ) = 0. (48)
ṙR (td ) = 0 (38)
Furthermore, by setting ẍLa (t0 + Ts ) = ẍRa (t0 + Ts ) =
r̈R (td ) < 0. (39) ÿLa (t0 + Ts ) = ÿRa (t0 + Ts ) = 0, the inequalities (45) and (46)
Under the above constraints and using (32), (33), (35), and can be rewritten as
(36), the boundary conditions of the x and y components of each Ls
ÿh (t0 + Ts ) < − |ẍh (t0 + Ts )| . (49)
joint’s position at the end of the double support phase can be 4yh (t0 + Ts )
obtained as in Tables 1 and 2. Furthermore, the inequality (39)
A proper choice for ẍh and ÿh is proposed as
can be rewritten as
ẍh (t0 + Ts ) = 0 (50)
ẍh (td ) (xh (td ) − xRa0 ) + ÿh (td )yh (td ) + ẋ2h (td ) + ẏ2h (td ) < 0. (40)
2Ls
It can be proved that the defined boundary conditions satisfy ÿh (t0 + Ts ) = − . (51)
Ts2
the inequality (40).
The boundary conditions of the joints’ positions for the single
4.2. Single support phase support phase are summarized in Tables 3 and 4.
The real-time walking pattern generation method is developed
In the single support phase, the right leg swings from xRa0 using the proposed real-time trajectory planners Tx and Ty , and
to xLa0 + 0.5Ls , while the left leg rests on xLa0 . Regarding to the the determined endpoint boundary conditions. By calculating xR ,
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 9

Fig. 12. Trajectory planned by Tx for fixed boundary conditions. Dashed lines indicate the boundary conditions. Solid lines indicate the planned trajectory, the first
and the second derivatives, and the control signal.

Fig. 13. Trajectory planned by Tx for changeable boundary conditions. Dashed lines indicate the boundary conditions. Solid lines indicate the planned trajectory, the
first and the second derivatives, and the control signal.

Table 4 5. Experimental results

Boundary conditions of the y component of each joints position for the single
support phase (tf = t0 + Ts ).
5.1. Simulation of the designed trajectory planners
Joint Hip Left ankle Right ankle
√
yj (tf ) l2 − (0.25Ls )2 0 0 The designed trajectory planners Tx and Ty play the main roles
ẏj (tf ) 0 0 0 in the walking pattern generation and have an effect on the
ÿj (tf ) − 2L
T2
s
0 0 performance of the gait. Therefore a performance analysis of the
s
p
yj l 0 Hs trajectory planners is required, especially for their response to
the changing boundary conditions. Resulting from the simulation,
Figs. 12 and 13 show the trajectories planned by Tx , for fixed and
changeable boundary conditions respectively. Fig. 12 shows the
yR , xL , and yL from (2) and applying inverse kinematics given by generated trajectory starting from the initial condition Xj (t0 ) =
(3), the joint angles will be obtained. [0 0 0]T and ending up at the endpoint boundary condition
10 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 14. Trajectory planned by Ty for fixed boundary conditions. Dashed lines indicate the boundary conditions. Dot-dashed line indicates the peak value. Solid lines
indicate the planned trajectory, the first and the second derivatives, and the control signal.

Fig. 15. Trajectory planned by Ty for changeable boundary conditions. Dashed lines indicate the boundary conditions. Dot-dashed line indicates the peak value. Solid
lines indicate the planned trajectory, the first and the second derivatives, and the control signal.

Xj (tf ) = [2 1 1]T at tf = 5. While in Fig. 13, the boundary the figures, the continuity of the first and the second derivatives
conditions change from Xj (tf ) = [2 1 1]T to Xj (tf ) = [1 − are maintained against the change of the parameters. As well
0.5 − 1]T at t = 2 and the boundary time is brought forward as, the final conditions of the trajectories reach to the required
from tf = 5 to tf = 4 at t = 3. Similarly for Ty , Figs. 14 and 15 boundaries at t = tf . The results illustrate the good performance
represent the trajectories, respectively for fixed and changeable of the trajectory planners.
boundary conditions. Fig. 14 shows the trajectory generated by
p
Ty for Yj (t0 ) = [0 1 0]T , Yj (tf ) = [1 − 1 − 1]T , yj = 2, and tf = 5. 5.2. The real-time walking pattern implementation
While in Fig. 15, the boundary parameters change three times at
t = 1, 2, and 3. In the figures, the dashed lines are the boundary For evaluating the proposed walking pattern generation
conditions, while the solid lines indicate the planned trajectory, method, different experiments with different walking parameters
the first and the second derivatives of the trajectory, and the were carried out on the Exoped robot. In order to evaluate the
control signal of the trajectory planner systems. As shown in robustness of the method across different users, each experiment
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 11

Fig. 16. Walking parameters of the first experiment.

Fig. 17. The x component of the position of the joints in the first experiment.

Table 5 the walking ends with a half-step taken by the left leg. Fig. 16
The characteristics of the participants.
shows the walking parameters related to the first experiment,
Participant lT lS Weight
which took 20 s. As shown, the half-step and full-step time was
User1 45 cm 51 cm 82 kg
User2 42 cm 47 cm 71 kg
2 and 4 s respectively. Maximum foot clearance and step length
were considered as Hs = 10 cm and Ls = 60 cm. The zero
step length from t = 18 s to t = 20 s corresponds to the final
half-step.
was repeated by two participants described in Table 5. The
Figs. 17 and 18 show the real-time position of the joints
experimental results of user1 are presented in the following. As
well as, the effect of the participant’s dimensions on the backward generated by the proposed method for the first experiment. As
balance of the robot is studied by comparing the stability results shown by the figures, the required walking parameters were
of the users. satisfied. Moreover, all of the trajectories had a continuous second
derivative.
5.2.1. The 1st experiment By applying inverse kinematics transformations, the desired
In the first experiment, walking begins from a standing pose angle trajectories of the joints were obtained in real-time as
and the right leg takes the first half-step. After taking four strides, shown in Fig. 19. A PID controller was used for regulation of
12 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 18. The y component of the position of the joints in the first experiment.

Fig. 19. Experimental result; The reference angles (black lines) and the real angles (red lines) of the joints in the first experiment . (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

each motor’s reference input. The obtained reference angles of the 5.2.2. The 2nd experiment
joints compared to the real angles measured by the hall sensors The second experiment was carried out to evaluate the efficacy
of the proposed method in obstacle crossing. As shown in Fig. 22,
are depicted in the plots. Also, the motor current corresponding
maximum foot clearance increased at t = 5 s from Hs = 8 cm to
to each joint is shown in Fig. 20.
Hs = 15 cm.
Fig. 21 Presents snapshots of whole walking cycle during the Figs. 23 and 24 show the generated real-time position of the
first experiment. joints. As shown in Fig. 24, the height of the stride increased from
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 13

Fig. 20. Experimental result; The motor currents of the joints in the first experiment.

Fig. 21. Snapshots of whole walking cycle during the first experiment.

Fig. 22. Walking parameters of the 2nd experiment.

14 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 23. The x component of the position of the joints in the 2nd experiment.

Fig. 24. The y component of the position of the joints in the 2nd experiment.

8 cm to 15 cm in the second full-step. In addition, the continuity 5.2.3. The 3rd experiment
of the second derivative of the trajectories are maintained in In the 3rd experiment the step length was shortened at the
respect to the change of the parameters at t = 5 s and t = 10 s. beginning of the second full-step and then reset back to the
Figs. 25 and 26 show the implementation results of the second previous value at the middle of the third step. Fig. 27 shows the
experiment. walking parameters during the experiment.
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 15

Fig. 25. Experimental result; The reference angles (black lines) and the real angles (red lines) of the joints in the 2nd experiment . (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 26. Experimental result; The motor currents of the joints in the 2nd experiment.

Fig. 27. Walking parameters of the 3rd experiment.

16 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 28. The x component of the position of the joints in the 3rd experiment.

Fig. 29. The y component of the position of the joints in the 3rd experiment.

The generated real-time positions of the joints are shown in 5.2.4. The 4th experiment
Figs. 28 and 29. As shown by the figures, the required step length The real-time changing of the walking speed is implemented
was satisfied in addition to maintaining continuity of the second in the 4th experiment. As shown in Fig. 32 the walking speed
derivatives. Figs. 30 and 31 show the implementation results of decreased at t = 6.2 s and then reset back to the previous value
the experiment. at t = 12 s.
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 17

Fig. 30. Experimental result; The reference angles (black lines) and the real angles (red lines) of the joints in the 3rd experiment . (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 31. Experimental result; The motor currents of the joints in the 3rd experiment.

Fig. 32. Walking parameters of the 4th experiment.

18 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 33. The x component of the position of the joints in the 4th experiment.

Fig. 34. The y component of the position of the joints in the 5th experiment.

Figs. 33 and 34 show the generated real-time position of the 5.2.5. The 5th experiment
joints. As shown by the figures, the required walking parameters The 5th experiment was carried out to evaluate the efficacy
were satisfied in addition to maintaining continuity of the second of the proposed method in response to the changes in all the
derivatives. Figs. 35 and 36 show the implementation results of parameters. Fig. 37 shows the walking parameters during the
the 4th experiment. experiment. As shown, maximum foot clearance and step length
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 19

Fig. 35. Experimental result; The reference angles (black lines) and the real angles (red lines) of the joints in the 4th experiment . (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 36. Experimental result; The motor currents of the joints in the 4th experiment.

Table 6 decision to put the left leg down 20 cm ahead of the right leg
Comparing the stability of the gaits for user1 and user2. in 1.2 s. This corresponds to setting Ls = 40 cm and Ts =
SM(t0 + 0.2Ts ) 2.7 s.
∫
Experiment Participant SM(t)dt
User1 5.029 0.0312 Figs. 38 and 39 show the generated real-time position of the
Exp1
User2 5.139 0.0326 joints. As shown by the figures, the required walking parameters
User1 3.981 0.0314 were satisfied in addition to maintaining continuity of the second
Exp2
User2 4.067 0.0330 derivatives. Figs. 40 and 41 show the implementation results of
User1 3.534 0.0290 the 5th experiment.
Exp3
User2 3.609 0.0302
User1 4.254 0.0278 5.2.6. Stability analysis
Exp4
User2 4.338 0.0288 By defining the stability margin SM in the single support phase
User1 3.005 0.0279 as
Exp5
User2 3.066 0.0291
SM (t ) = ẋh (t ) − ω(max(xRa (t), xLa (t)) − xh (t)), (52)
the following constraint should be satisfied to avoid a back-
ward balance loss during the single support phase
increased at t = 1.8 s and t = 6 s respectively. Also, walk-
ing speed decreased at t = 4.5 s. The walking ends with a SM (t ) > 0. (53)
20 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 37. Walking parameters of the 5th experiment.

Fig. 38. The x component of the position of the joints in the 5th experiment.

The value of SM(t) for each experiment is given in Fig. 42, results for user1 and user2, it can be seen that the ∼7% decrease
in which SM (t ) is positive for all experiments and the stability in l = lT + lS leads to ∼4% improvement in stability of the gaits
of the gaits are preserved. By comparing SM(t) in the first and at the critical time (beginning of the double support phases) and
second experiment, it can be said that in the proposed method ∼2% improvement in stability of the gaits during the walking
the backward balance is not affected by Hs . While, the 4th exper- cycles.
iment (6.2 < t < 12) demonstrates that a significant decrease in
walking speed can lead to backward balance loss. 6. Conclusion
The stability margin obtained for user1 and user2 is compared
in order to investigate the effect of the user dimensions on the A walking pattern generation method is proposed which gen-
stability of the gaits. For this purpose, the integrals of SM (t ) erates the joint angles based on gait parameters. The provided
during the experiments, and the mean values of SM(t0 + 0.2Ts ) pattern generator is designed in such a way that enables changing
at the start time of the double support phases are computed. The the gait parameters during the stride. For this purpose, the real-
obtained values are summarized in Table 6. By comparing the time trajectory planning problem is formulated as an optimal
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 21

Fig. 39. The y component of the position of the joints in the 5th experiment.

Fig. 40. Experimental result; The reference angles (black lines) and the real angles (red lines) of the joints in the 5th experiment . (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)

control problem with changeable final states; and a novel method Exoped robot. The results indicate that the robot has a smooth
is proposed to solve this problem. The trajectories generated by and stable motion even by changing the walking parameters.
the proposed method have continuous second derivatives, even The proposed method provides a natural walking pattern for
with the changing parameters of gait. In order to achieve a stable the lower limb exoskeletons, which is adaptable to the change
walking gait, the trajectory of the center of mass is defined in of the walking parameters in different situations and moments.
such a way to prevent the backward balance loss. Two experi- Moreover, the proposed method provides a safe way for walking
ments were carried out to evaluate the proposed method on the with crutches by maintaining the backward balance. Considering
22 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23

Fig. 41. Experimental result; The motor currents of the joints in the 5th experiment.

Fig. 42. The value of SM(t) during the experiments for user1. SM(t) is defined as zero in double support phases.

the stability in the gait can help the user by reducing the force References
applied to the arms, especially in the clinical training between the
[1] R.S. Mosher, Handyman to Hardiman, SAE Trans. 76 (1968) 588–597.
parallel bars. According to Fig. 3 our future work is to design the
[2] A. Formalskii, A. Shneider, Book Reviews : Biped Locomotion (Dynamics
walking parameters generator block. As an example, the sensor Stability Control and Application): by M. Vukobratovich, B. Borovac, D.
based determination of the parameters can be used for automat- Surla, and D. Stokich published by Springer-Verlag, 1990, Int. J. Robot. Res
11 (4) (1992) 396.
ing the movement of the robot in order to overcome obstacles [3] E. Guizzo, H. Goldstein, The rise of the body bots [robotic exoskeletons],
and difficult situations. As well as, the real-time computing of IEEE Spectr. 42 (10) (2005) 50–56.
the walking parameters according to the stability of the robot [4] T. Yan, M. Cempini, C.M. Oddo, N. Vitiello, Review of assistive strategies
in powered lower-limb orthoses and exoskeletons, Robot. Auton. Syst. 64
can make the movement of the robot safer. For this purpose, the (2015) 120–136.
foot force stability margin [19] is a simple and reliable index for [5] M. Vukobratović, B. Borovac, Zero-moment point — thirty five years of its
assessing stability of a exoskeleton robot. In any case, the real- life, Int. J. Humanoid Robot. 01 (01) (2004) 157–173.
[6] H.F.N. Al-Shuka, B. Corves, W.-H. Zhu, B. Vanderborght, Multi-level control
time determination of the walking parameters will be a potential of zero-moment point-based humanoid biped robots: a review, Robotica
application of the proposed walking pattern generator. 34 (11) (2016) 2440–2466.
 J. Kazemi and S. Ozgoli / Robotics and Autonomous Systems 116 (2019) 1–23 23

[7] H. Kazerooni, J.L. Racine, L. Huang, R. Steger, On the Control of the [17] D.E. Kirk, Optimal Control Theory: An Introduction, Courier Corporation,
Berkeley Lower Extremity Exoskeleton (BLEEX), in: Proceedings of the 2012.
IEEE 2005 International Conference on Robotics and Automation, 2005, [18] Waijung Blockset. [Online]. Available: http://waijung.aimagin.com/ [Ac-
pp. 4353–4360. cessed: 05-Jul-2018].
[8] C. Lee, J.-Y. Kim, S.-Y. Kim, S. Oh, Human force observation and assistance [19] M. Agheli, S.S. Nestinger, Force-based stability margin for multi-legged
for lower limb rehabilitation using wire-driven series elastic actuator, robots, Robot. Auton. Syst. 83 (2016) 138–149.
Mechatronics 55 (2018) 13–26.
[9] A. Esquenazi, M. Talaty, A. Packel, M. Saulino, The rewalk powered
exoskeleton to restore ambulatory function to individuals with thoracic-
level motor-complete spinal cord injury, Am. J. Phys. Med. Rehabil. 91 (11) Jafar Kazemi received his B.Sc. in electrical engineering
(2012) 911–921. from Tabriz University, Tabriz, Iran, in 2012, and his
[10] K.A. Strausser, H. Kazerooni, The development and testing of a human M.Sc. in electrical engineering from Tarbiat Modares
machine interface for a mobile medical exoskeleton, in: 2011 IEEE/RSJ University, Tehran, Iran, in 2014. He is currently a Ph.D.
International Conference on Intelligent Robots and Systems, 2011, pp. student in the Electrical and Computer Engineering
4911–4916. Department at Tarbiat Modares University, Tehran, Iran.
[11] D. Sanz-Merodio, M. Cestari, J.C. Arevalo, E. Garcia, Control motion ap- His current research interest focuses on the lower limb
proach of a lower limb orthosis to reduce energy consumption, Int. J. Adv. exoskeletons.
Robot. Syst. 9 (6) (2012) 232.
[12] J. Yoon, R.P. Kumar, A. Özer, An adaptive foot device for increased gait and
postural stability in lower limb Orthoses and exoskeletons, Int. J. Control
Autom. Syst. 9 (3) (2011) 515. Sadjaad Ozgoli is an associate professor working with
[13] R. Huang, H. Cheng, Y. Chen, Q. Chen, X. Lin, J. Qiu, Optimisation of the electrical engineering school at Tarbiat Modares
reference gait trajectory of a lower limb exoskeleton, Int. J. Soc. Robot. University. He works on control systems with appli-
8 (2) (2016) 223–235. cation to Mechatronics systems and Biological systems.
[14] F.M. Silva, J.A.T. Machado, Kinematic aspects of robotic biped locomotion He received his B.Sc. and M. Eng degrees in electri-
systems, in: Proceedings of the IEEE/RSJ International Conference on cal engineering from Sharif University of Technology,
Intelligent Robots and Systems, 1997. IROS ’97, 1997, pp. 266–272 vol.1. Tehran, Iran, respectively in 1997 and 1999. He
[15] E.T. Esfahani, M.H. Elahinia, Stable walking pattern for an SMA-actuated received his Ph.D. from K.N. Toosi University of Tech-
biped, IEEE ASME Trans. Mechatron. 12 (5) (2007) 534–541. nology, Tehran, Iran in electrical engineering with
[16] T. Kagawa, H. Ishikawa, T. Kato, C. Sung, Y. Uno, Optimization- based minors in mechanical engineering in 2005. He has pub-
motion planning in joint space for walking assistance with wearable robot, lished more than 40 papers in international Journals
IEEE Trans. Robot. 31 (2) (2015) 415–424. and conference proceedings.