IEEE PEDS 2023, Montreal, Canada

7 – 10 August 2023

Robust Fast Terminal SMC with Prescribed

Performance for a Wearable Exoskeleton Robot
2023 IEEE 14th International Conference on Power Electronics and Drive Systems (PEDS) | 979-8-3503-9996-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/PEDS57185.2023.10246747

Mahmoud Abdallah1 , Yassine Kali1 , Maarouf Saad1 , Raouf Fareh2 and Maamar Bettayeb2,3
1
École de Technologie Supérieure, Montreal, QC H3C 1K3, Canada
mahmoud-a-y.abdallah.1@ens.etsmtl.ca; {yassine.kali; maarouf.saad}@etsmtl.ca
2
University of Sharjah, Sharjah, UAE
{rfareh; maamar}@sharjah.ac.ae
3
CEIES , King Abdulaziz University , Jeddah , KSA

Abstract—This document proposes a robust fast terminal sliding surface. SMC has been successfully applied to various
sliding mode control system with prescribed performance for MIMO robotic systems [8]–[11]. However, the main drawback
an upper limb exoskeleton robot used for rehabilitation pur- of SMC is the chattering phenomenon. The chattering couldn’t
poses. The controller is introduced to achieve highly accurate
performance for trajectory tracking in the joint space under the just damage the actuators of a robotic system, but also produce
effect of both internal uncertainties and external perturbations. high vibrations that could significantly affect the rehabilitation
The developed prescribed performance function plays the main process quality.
role in converging the tracking error to a user-defined small To overcome the problem caused by the chattering phe-
gap rapidly. The expression of the conventional terminal sliding nomena in exoskeletons, a SMC with model-based switching
mode control system has been rewritten to meet the prescribed
function effect. A stability analysis based on Lyapunov theory functions has been applied to a 7-DoF exoskeleton robot [12].
has been delivered to ensure that the system is stable. A The experimental results showed that the introduced model-
simulation experiment on the 7-DoF ETS-MARSE robot and based switching functions are able to provide more robustness
a comparison with conventional SMC have been conducted to and less tracking error when compared with conventional lin-
prove the effectiveness and robustness of the proposed control ear switching functions-based SMC. However, the convergence
scheme.
Index Terms—Exoskeleton, tracking control, torque, rehabili-
time and chattering could be reduced.
tation, prescribed performance An alternative solution to the chattering problem is the
Super-Twisting Algorithm (STA) with Time Delay Estimation
I. I NTRODUCTION (TDE) [13], [14]. The control scheme uses the TDE to estimate
the uncertainties and feeds these estimated parts to the STA for
As the number of stroke victims increases, the demand the aim of robustness, finite-time convergence, and chattering
for optimal rehabilitation therapies also increases. Exploiting reduction. The work showed an advantage of the proposed
robotic systems in rehabilitation therapy is a leading therapy controller over standard STA and conventional SMC. However,
solution that attracts collaboration between robotics developers the control scheme deals with a class of matched uncertainties
and therapists. Rehabilitation centers are utilizing exoskeleton only while the exoskeleton is subject to a larger class of
robots to help rehab patients with upper-limb disabilities [1]. uncertainties and disturbances.
Exoskeletons are a special type of rigid serial-link manipulator In this paper, a robust nonlinear control scheme with pre-
that allows the user to fit and tie his upper limb with its scribed performance consideration for wearable exoskeleton
outer body. Although exoskeletons are used for rehabilitation, robots is presented. The main role of this controller is to
they require a control system that ensures efficient and secure achieve robustness in trajectory-tracking scenarios under the
tracking. passive rehabilitation process. The first step in the controller
Many control systems have been introduced for serial-link design is to define the prescribed performance function which
manipulators and exoskeleton robots especially [2]. Despite limits the tracking error in joint space, decreases with time,
that a linear control system such as conventional PID can be and converges. This will ensure the evolution of the joint
used for exoskeleton robots [5], it fails to meet robustness tracking vector inside a canal. After that, a virtual control
against matched uncertainties in real-time implementation. input that is a function of the new tracking error is defined.
Mainly, non-linear control systems are utilized for high-Degree Finally, a non-singular terminal sliding surface is defined. A
of Freedom (DoF) exoskeletons. These control systems are Lyapunov function has been considered to prove the stability
model-dependent techniques that require a pre-known dynamic of the proposed sliding surface and thus make sure that the
model of the robot [3], [4]. They aim to reduce the tracking er- dynamics of the robot are stable and that there is a globally
ror and stabilize the system. One of the well-known nonlinear stable origin. The proposed controller has been simulated on
control systems used in exoskeletons is the Sliding Mode Con- a 7-DoF exoskeleton robot (ETS-MARSE) dynamic model
trol (SMC) [7]. SMC tends to reduce the order of nonlinearity which will be introduced in Section 2. The control scheme
of the robot dynamics by deriving the systems variable on the reduces the effect of the chattering phenomena and has a faster

Authorized licensed use limited to: Bibliothèque ÉTS. Downloaded on November 09,2023 at 16:43:36 UTC from IEEE Xplore. Restrictions apply.
convergence time, better steady-state error, and as well as a
prescribed performance.
With the introduction presented in this section, the remain-
ing sections of this paper are: Section II is the robot model
and problem formulation; Section III is the controller design;
Section IV is the case study, and Section V is the conclusion.
II. ROBOT M ODEL AND P ROBLEM FORMULATION
A. ETS-MARSE Description
The ETS-MARSE [12] is a seven-DoF robot prototype that
allows performing a variety of passive, assisted, and active
exercise tasks on the impaired upper limb. Each joint is limited
within a range of motion to provide safety to the wearer
as physical limits shown in Table I. Figure 1 shows the 7
joints with the assigned frames according to the modified
Denavit-Hartenberg (D-H) convention [12]. The modified D-H
parameters can be found in [12]. The equation of motion of
the 7-DOF ETS-MARSE can be expressed as follows:
D(θ)θ̈ + C(θ, θ̇)θ̇ + G(θ) = τ + τp (1)
where θ ∈ R7 is the joint position vector, D(θ) ∈ R7×7 is
the non-singular and bounded inertia matrix, C(θ, θ̇)θ̇ ∈ R7 is
the vector of centrifugal and Coriolis forces, G(θ) ∈ R7 is the Fig. 1. Link frame attachment [15]
gravitational vector, τp ∈ R7 is the vector of perturbations
and finally τ ∈ R7 is the torques input vector. Taking
into consideration the presence of uncertainties, (1) can be B. Problem Formulation
rewritten as:
h i The main objective of robotic exoskeleton control systems
θ̈ =D0−1 (θ) τ − F (θ, θ̇) + N (θ, θ̇, θ̈) is to ensure tracking in joint space and thus the tracking in
(2) the workspace. This objective should lead to a minimized
F (θ, θ̇) =C0 (θ, θ̇)θ̇ + G0 (θ)
tracking error and a stable system under operation limitations
where D0 (θ) ∈ R7×7 , C0 (θ, θ̇)θ̇ ∈ R7 and G0 (θ) ∈ R7 are while ensuring safe rehabilitation exercises. The objective is
respectively the known elements of D(θ), C(θ, θ̇)θ̇ and G(θ) expressed in the following equation:
while N (θ, θ̇, θ̈) is defined by:
lim |θ̃i (t)| = 0 (5)
h i t→∞
N (θ, θ̇, θ̈) = −D0−1 (θ) Du (θ)θ̈ + Cu (θ, θ̇)θ̇ + Gu (θ) − τp (3)
III. C ONTROLLER DESIGN
where Du (θ), Cu (θ, θ̇) ∈ R7×7 and Gu (θ) ∈ R7 are
This part aims to propose a new sliding manifold that
respectively the unknown elements of D(θ), C(θ, θ̇) and G(θ).
ensures fixed-time convergence while improving the tracking
In the following, it is assumed that the function of the unknown
performance by imposing constraints on the trajectory tracking
vector N (θ, θ̇, θ̈) = [N1 , · · · , N7 ]T is continuous and bounded
error. To that end, let us specify the following limits [16] for
such as for i = 1, · · · , 7:
i = 1, · · · , 7:
|Ni | ≤ N i (4)
−δi ϕi (t) < θ̃i = θi − θid < δi ϕi (t) (6)
where N i is a known positive constant.
where δi ∈ (0, 1) and ϕi (t) is the prescribed performance
TABLE I function defined as follows:
R ANGE OF M OTION OF THE ETS-MARSE
ϕi (t) = ϕfi − (ϕfi − ϕ0i )exp(−αi t) (7)
Joint The effect of its rotation Rotation range
1 Shoulder flexion & extension 140◦ , 0◦
2 Shoulder abduction & adduction 140◦ , 0◦ where 0 < ϕfi < ϕ0i and αi is a positive constant for i =
3 Shoulder internal & external rotation −85◦ , 75◦ 1, · · · , 7. The above performance function is designed to be a
4 Elbow and forearm flexion & extension 120◦ , 0◦ decreasing function such as ϕi (t) ≥ 0 and ϕ̇i (t) ≤ 0 ∀t, and
5 Elbow and forearm pronation & supination −85◦ , 85◦
6 Wrist flexion & extension 60◦ , 50◦
converges to ϕfi such as limt→∞ ϕi (t) = ϕfi > 0.
7 Wrist radial & ulnar deviation 20◦ , −25◦ Now, let us introduce a new variable ψi such as:

θ̃i = ϕi (t)ψi (8)

Authorized licensed use limited to: Bibliothèque ÉTS. Downloaded on November 09,2023 at 16:43:36 UTC from IEEE Xplore. Restrictions apply.
 h iT
where: • θ̈d = θ̈1d , · · · , θ̈7d is the desired acceleration vector,
exp(εi ) − exp (−εi )
 
ψi = δ i (9) • Φ = diag(ϕ1 , · · · , ϕ7 ),
exp(εi ) + exp(−εi ) f
• Φ̇ = diag(ϕ̇1 , · · · , ϕ̇7 ) with ϕ̇i = αi (ϕi − ϕi )exp
0 (−αi t)
,
2 0 f (−αi t)
with εi represents the transformed tracking error given by: • Φ̈ = diag(ϕ̈1 , · · · , ϕ̈7 ) with ϕ̈i = αi (ϕi − ϕi )exp ,
!0.5 δi
• R = diag(r1 , · · · , r7 ) with ri = ,
δi ϕi (t) + θ̃i δi2 ϕ2i (t) − θ̃i2
εi = ln (10)
δi ϕi (t) − θ̃i ˙
2δi θ̃i θ̃i − 2δi3 ϕi ϕ̇i
• Ṙ = diag(ṙ1 , · · · , ṙ7 ) with ṙi =  2 ,
Taking the first time derivative of the above transformed error δi2 ϕ2i (t) − θ̃i2
variable gives:
= diag |ε1 |γ1 −1 , |ε2 |γ2 −1 , · · · , |ε7 |γ7 −1 ,
γ−1


• ⌊ε⌋
˙ • Γ = diag(γ1 , · · · , γ7 ).
δi ϕi (t)θ̃i − δi ϕ̇i (t)θ̃i
ε̇i = (11)
δi2 ϕ2i (t) − θ̃i2 The above equation is used to compute the equivalent control
by resolving ṡ = 0:
Based on the transformed error and its derivative, the
following fast terminal sliding surface is designed as: τeq = D0 (θ)θ̈d − D0 (θ)Φ−1 R−1 τ0 + F (θ, θ̇) (19)

s = ε̇ + Kε + Λsigγ (ε) (12) where:

˙
 
τ0 = Ṙ Φθ̃ − Φ̇θ̃ − RΦ̈θ̃ + K ε̇ + ΛΓsat ⌊ε⌋γ−1 ε̇, µ (20)


where K = diag(k1 , · · · , k7 ) and Λ = diag(λ1 , · · · , λ7 )
are diagonal positive-definite matrices, sigγ (ε) = The saturation function (sat) is used in the above equation
 γ T
|ε1 | 1 sign(ε1 ), . . . , |ε7 |γ7 sign(ε7 ) with γi is defined to avoid the singularity problem that might occur during the
for i = 1, · · · , 7 by: ˙

sliding phase when θ̃i = 0 and θ̃i ̸= 0 and sat ⌊ε⌋γ−1 ε̇, µ =

T
sat |ε1 |γ1 −1 ε̇1 , µ1 , · · · , sat |ε7 |γ7 −1 ε̇7 , µ7


1 + γui , if |εi | > 1 such as:
γi = (13)
1 − γli , if |εi | ≤ 1 
⋆i , if |⋆i | < µi
sat (⋆i , µi ) = (21)
with γui > 0 and γli ∈ (0, 0.5). Otherwise, the signum µi sign(⋆i ), if |⋆i | ≥ µi
function is defined by:
 with ⋆i = |εi |γi −1 ε̇i for i = 1, · · · , 7. Since the robot model is
 −1, if εi < 0 not completely known because of the parameter variations, the
sign(εi ) = 0, if εi = 0 (14) disturbances and the unknown human-robot interaction force,
the following switching action is added to ensure robustness:
1, if εi > 0

τsw = −D0 Φ−1 R−1 [β1 s + β2 sigν1 (s) + β3 sigν2 (s) + β4 sign(s)]
The above proposed surface is stable at εi = 0 (i.e., θ̃i = 0) (22)
for i = 1, · · · , 7. To prove it, let us consider the following where βj are (7 × 7) diagonal positive definite matrices for
positive-definite Lyapunov function and its first time deriva- j = 1, 2, 3, 4 chosen to satisfy the closed-loop stability and
T
tive: sigνm (s) = [|s1 |νm1 sign(s1 ), · · · , |s7 |νm7 sign(s7 )] for m =
V1 =εT ε 1, 2 with ν1i > 1 and 0.5 < ν2i < 1.
(15)
V̇1 =2εT ε̇ Finally, the proposed controller is obtained by combining
both the equivalent (19) and the switching (22) actions as:
During the sliding phase, s = 0. Then:
τ = τeq + τsw (23)
ε̇ = −Kε − Λsigγ (ε) (16)
To study the stability of the overall system, let us consider
Substituting the above equation into V̇1 yields to: the following positive-definite Lyapunov function and its first
V̇1 = 2εT (−Kε − Λsigγ (ε)) time derivative:
V2 =0.5 sT s
7
(17) (24)
V̇2 =sT ṡ
X
=− 2ki ε2i + 2λi |εi |1+γi ≤ 0
i=1 where ṡ is the closed-loop error dynamics obtained by sub-
Now, based on (11) and (12) and using the nominal robot stituting the designed controller (23) into the model of the
dynamics in (2), the first time derivative of the proposed wearable robot arm (2). Hence:
sliding surface s is computed as: V̇2 =sT RΦN − β1 s − β2 sigν1 (s) − β3 sigν2 (s) − β4 sign(s)
 

ṡ =ε̈ + K ε̇ + ΛΓ⌊ε⌋γ−1 ε̇ 7
X
¨

˙
 = ri ϕi Ni si − β1i s2i − β2i |si |ν1i +1 − β3i |si |ν2i +1 − β4i |si |
=RΦθ̃ − RΦ̈θ̃ + Ṙ Φθ̃ − Φ̇ + K ε̇ + ΛΓ⌊ε⌋γ−1 ε̇ i=1
(18) 7
˙
h i  
=RΦD0−1 (θ) τ − F (θ, θ̇) − RΦθ̈d + Ṙ Φθ̃ − Φ̇θ̃
X
N i |ri ϕi | − β4i |si | − β1i s2i − β2i |si |ν1i +1 . . .


≤
i=1
− RΦ̈θ̃ + K ε̇ + ΛΓ⌊ε⌋γ−1 ε̇
. . . − β3i |si |ν2i +1
where: (25)

Authorized licensed use limited to: Bibliothèque ÉTS. Downloaded on November 09,2023 at 16:43:36 UTC from IEEE Xplore. Restrictions apply.
 The above V̇2 is negative definite if the switching gains β4i
are chosen to satisfy the following condition:
β4i ≥ N i |ri ϕi | (26)
IV. N UMERICAL S IMULATION
In this section, a simulation is conducted on the mathemat-
ical model of the ETS-MARSE. This simulation aims to test
and validate the designed controller introduced in Section 3.
The ETS-MARSE robot is assigned a passive rehabilitation
task to track a predefined path in the workspace. To verify the
effectiveness of the proposed controller, a comparison study
with the conventional SMC is done. Based on trial and error,
the gains of the prescribed fast terminal SMC are set to:
K =diag7×7 (150), Λ = diag7×7 (100), ϕfi = 0.01,
ϕ0i =0.1, δi = 1, αi = 2, µi = 0.3,
(27)
γui =0.2, γli = 0.4, ν1i = 1.7, ν2i = 0.7,
β1 =β2 = β3 = diag7×7 (1), β4 = diag7×7 (0.3)
The conventional SMC for robot manipulators is given by:
˙ ˙
  
τ = D0 (θ) θ̈d − β θ̃ − M sign θ̃ + β θ̃ + F (θ, θ̇) (28)

where the gains are set to:

M = diag7×7 (0.1), β = diag7×7 (20) (29)

Fig. 3. Tracking in joint space

TABLE II
MSE IN WORKSPACE ( METER )

MSE X-axis MSE Y-axis MSE Z-axis

Proposed controller 2.7 × 10−6 3.7 × 10−6 5.1 × 10−6
Conventional SMC 8.3 × 10−5 8.03 × 10−4 4.2 × 10−4

Table II shows the mean-square error (MSE) in the

workspace of both the prescribed performance-based fast
terminal SMC and the conventional SMC.
Table II showed that the MSE of tracking in the workspace
using the prescribed fast terminal SMC controller has less
value than the tracking using the conventional SMC controller.
As shown in Fig. 2, the task was to follow a rectangular
trajectory in the 3D workspace for a period of 8 seconds. It is
clearly noticed that the trajectory performed by the prescribed
SMC is almost overlapping the desired trajectory, where the
trajectory performed by the conventional SMC is shifted on
the X-axis and the Z-axis.
In Fig. 3, the tracking performance for each joint of both the
prescribed fast terminal SMC and the conventional SMC are
shown. The first 4 joints in the ETS-MARSE are responsible
Fig. 2. Workspace tracking
to change the position of the end-effector. Thus, the good
tracking of joints 1 and 2 in the prescribed performance are

Authorized licensed use limited to: Bibliothèque ÉTS. Downloaded on November 09,2023 at 16:43:36 UTC from IEEE Xplore. Restrictions apply.
 Fig. 4. Tracking error in joint space Fig. 5. Control torque inputs

the reason behind the good tracking in the workspace. It is effectiveness and robustness of the proposed control scheme
noticeable that conventional SMC has less tracking accuracy over the conventional SMC.
for these two joints (almost 0.1 rad difference from the desired
trajectory for both θ1 and θ2 tracking error).
R EFERENCES
The effect of the prescribed performance function defined
in (7) could be clearly observed from the tracking error for [1] Siddique, T., Fareh, R., Abdallah, M., Ahmed, Z. & Rahman, M.
each joint in Fig. 4. It is obvious that the prescribed SMC Autonomous Exercise Generator for Upper Extremity Rehabilitation: A
tracking error has faster convergence and less steady-state Fuzzy-Logic-Based Approach. Micromachines. 13, 842 (2022)
[2] Gull, M., Bai, S. & Bak, T. A review on design of upper limb
error than the one in the conventional SMC. Fig. 5 shows the exoskeletons. Robotics. 9, 16 (2020)
torque performed by each controller for each joint’s motor. [3] Kali, Y., Saad, M. & Benjelloun, K. Control of robot manipulators using
The prescribed SMC has less chattering that the conventional modified backstepping sliding mode. New Developments And Advances
SMC in all joints (noticeable in joints 3, 4, and 6). This is a In Robot Control. pp. 107-136 (2019)
[4] Abdallah, M. & Fareh, R. Fractional order active disturbance rejection
reason caused by the better steady-state error in the prescribed control for trajectory tracking for 4-DOF serial link manipulator. Inter-
SMC performance. national Journal Of Modelling, Identification And Control. 36, 57-65
(2020)
[5] Yu, W. Chapter 3 - PID Control in Task Space. PID Control With
V. C ONCLUSION Intelligent Compensation For Exoskeleton Robots. pp. 35-53 (2018)
[6] Rahman, M., Saad, M., Kenné, J. & Archambault, P. Control of an
This paper presented a robust fast terminal sliding mode exoskeleton robot arm with sliding mode exponential reaching law.
International Journal Of Control, Automation And Systems. 11 pp. 92-
control system with prescribed performance for an upper limb 104 (2013)
exoskeleton. The controller achieved an accurate performance [7] Young, K., Utkin, V. & Ozguner, U. A control engineer’s guide to sliding
for trajectory tracking in the joint space. The prescribed mode control. IEEE Transactions On Control Systems Technology. 7,
performance function played the main role in converging the 328-342 (1999)
[8] Yang, J. & Kim, J. Sliding mode control for trajectory tracking of
tracking error and having a better steady-state error. The nonholonomic wheeled mobile robots. IEEE Transactions On Robotics
simulation on the 7-DoF ETS-MARSE robot showed the And Automation. 15, 578-587 (1999)

Authorized licensed use limited to: Bibliothèque ÉTS. Downloaded on November 09,2023 at 16:43:36 UTC from IEEE Xplore. Restrictions apply.
 [9] Fareh, R., Saad, M. & Saad, M. Workspace tracking trajectory for
7-DOF ANAT robot using a hierarchical control strategy. 2012 20th
Mediterranean Conference On Control Automation (MED). pp. 122-
127 (2012)
[10] Rı́os, H., Falcón, R., González, O. & Dzul, A. Continuous Sliding-
Mode Control Strategies for Quadrotor Robust Tracking: Real-Time
Application. IEEE Transactions On Industrial Electronics. 66, 1264-
1272 (2019)
[11] Nguyen, X., Tran, T., Nguyen, H., Tong, N., Nguyen, T. & Nguyen, T.
Controlling Center of Mass in Humanoid Robot using Sliding Mode
Control. 2020 International Conference On Advanced Mechatronic
Systems (ICAMechS). pp. 17-22 (2020)
[12] Fallaha, C., Saad, M., Ghommam, J. & Kali, Y. Sliding mode control
with model-based switching functions applied on a 7-dof exoskeleton
arm. IEEE/ASME Transactions On Mechatronics. 26, 539-550 (2020)
[13] Kali, Y., Saad, M., Benjelloun, K. & Khairallah, C. Super-twisting
algorithm with time delay estimation for uncertain robot manipulators.
NONLINEAR DYNAMICS. 93, 557-569 (2018,7)
[14] Kali, Y., Ayala, M., Rodas, J., Saad, M., Doval-Gandoy, J., Gregor,
R. & Benjelloun, K. Time Delay Estimation Based Discrete-Time
Super-Twisting Current Control for a Six-Phase Induction Motor. IEEE
Transactions On Power Electronics. 35, 12570-12580 (2020)
[15] Brahmi, B., Saad, M., Rahman, M. & Ochoa-Luna, C. Cartesian trajec-
tory tracking of a 7-DOF exoskeleton robot based on human inverse
kinematics. IEEE Transactions On Systems, Man, And Cybernetics:
Systems. 49, 600-611 (2017)
[16] Jing, C., Xu, H. & Niu, X. Adaptive sliding mode disturbance rejection
control with prescribed performance for robotic manipulators. ISA
Transactions. 91 pp. 41-51 (2019)

Authorized licensed use limited to: Bibliothèque ÉTS. Downloaded on November 09,2023 at 16:43:36 UTC from IEEE Xplore. Restrictions apply.