International Journal of Control

ISSN: 0020-7179 (Print) 1366-5820 (Online) Journal homepage: https://www.tandfonline.com/loi/tcon20

ESO based sliding mode control for the welding

robot with backstepping

Pengcheng Wang, Dengfeng Zhang & Baochun Lu

To cite this article: Pengcheng Wang, Dengfeng Zhang & Baochun Lu (2020): ESO based
sliding mode control for the welding robot with backstepping, International Journal of Control, DOI:
10.1080/00207179.2020.1762932

To link to this article: https://doi.org/10.1080/00207179.2020.1762932

Accepted author version posted online: 29

Apr 2020.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tcon20
Publisher: Taylor & Francis & Informa UK Limited, trading as Taylor & Francis Group
Journal: International Journal of Control
DOI: 10.1080/00207179.2020.1762932

ESO based sliding mode control for the welding robot with
backstepping
Pengcheng Wang[1] Dengfeng Zhang[1] Baochun Lu[1]
[1]
Nanjing University of Science and Technology. Nanjing.China.
Corresponding author：1 Name: Wang Pengcheng, September 1989, Phd ; Institution: School of
Mechanical Engineering Nanjing University of Science and Technology. Nanjing.China;
Address:No.200,Hsiao Ling Wei, Xuanwu District, Nanjing, Jiangsu Province; Email:
[email protected]
Co-Corresponding author：2 Name: Zhang Dengfeng, October 1973, Doctoral tutor; Institution:
School of Mechanical Engineering Nanjing University of Science and Technology. Nanjing.China;
Address:No.200,Hsiao Ling Wei, Xuanwu District, Nanjing, Jiangsu Province; Email:
[email protected]
3 Name: Lu Baochun, June 1965, Doctoral tutor; Institution: School of Mechanical Engineering
Nanjing University of Science and Technology. Nanjing.China; Address: No.200,Hsiao Ling Wei,
Xuanwu District, Nanjing, Jiangsu Province; Email: [email protected]

Abstract：This paper investigates a problem of tracking control for robotic manipulators with high
accuracy. In view of the uncertainty and disturbance during the process of trajectory tracking,
ESO(Extended State Observer ) control is proposed to handle the uncertain factors(disturbance
and dynamic uncertainty) and estimate the external torque and kinematics, respectively. Based on
the combination of sliding mode control and backstepping control strategy, we design a new
controller , which has strong robustness and accuracy. In term of the dynamic uncertainty in the
process of trajectory tracking and considering the estimated information , the controller can
guarantee the desire trajectory and the error to be zero. Lyapunov function and barbalat theorem
are used to prove the global stability of the controller. During the simulation and experiment, the
results show the performance of the proposed controller and the results are analyzed and compared
with the PID controller. The error analysis is carried out to verify the performance of the method.
Key Words: ESO(Extended State Observer), sliding mode control, backstepping, robot,
robust, uncertainty.
1.Introduction
In recent years, the robots are playing more and more important role due to its self-delivery
control, so that people pay more and more attention to robots, especially in the ultra-high precision,
adequate perception and rapidly response capacity，and the demand for this is more and more high.
Considering the uncertainty of the measurement and modeling during the practical application, it
is impossible to get the accurate mathematical model without error, and the high quality trajectory
tracking control is more difficult to be realized.
In order to solve the uncertainty in trajectory tracking, the main dynamics control methods
are sliding mode control (SMC), robust control and adaptive control in the whole trajectory
tracking process. Due to the uncertainty in the actual control process and the phenomenon of
"chattering" existing in the SMC method control, SMC control and ESO and other intelligent
control methods often are combined to achieve better quality, which is now a common trend (Tang,
1998; Guo and Woo, 2003; Park and Lee, 1997; ; Ginoya et al., 2015; Chen, 2001),which has
carried on the theory and the application research to the related field(Qiao and Zhang, 2018; Qiao
and Zhang, 2017).
Sliding mode control is an effective and popular control strategy, especially for the dynamic
uncertainty and external disturbance in the control system. In the practical application, the
traditional SMC exhibit insensitive to the matching uncertainties and external disturbances
(Capisani and Ferrara, 2011; Fallaha et al., 2011; Ferrara and Incremona, 2015; Zhao et al., 2015;
Yagiz and Hacioglu, 2009). Combining the SMC method with other control methods to estimate
the uncertainties and external disturbances is a hot topic for modern scholars (Tang, 1998; Guo
and Woo, 2003; Park and Lee, 1997; ; Ginoya et al., 2015; Chen, 2001). In (Cui et al., 2017), a
new integral sliding mode controller is designed, based on multi-input and multi-output state
observer and the uncertainty of the mechanical dynamic uncertainty and disturbance. The
experimental research of underwater robot is carried out, and the performance of the algorithm is
verified. In (Madani and Benallegue, 2007), in view of the nonlinear robot system, a control based
on exponential arrival law is designed, and the simulation research is carried out, and the validity
of the related controller is proved. In (Lu et al.), the research on the under-driven control system is
carried out, the correlative simulation and experiment are carried out which the method of SMC
and state observer, which solves the problem of the matching and non-matching disturbance in the
control process of the crane. (Ginoya et al., 2015), (Tran and Kang, 2016), (Chan et al., 2013)
show the performance of the fast terminal sliding mode controller and verified the effectiveness of
the proposed controller through the experiments. In (Precup et al., 2017), allowing for the
uncertainty in the modeling of muscle torque, a nonlinear observer method is proposed and the
control strategy of the terminal sliding mode is put forward. Experimental results showed that the
method of terminal sliding mode control with observer which can solve the problem of modeling
error and external disturbance well.
Backstepping control is a non-linear system design method , and it introduces the virtual control,
decomposing the complex nonlinear system into more simple and order systems. This approach is
often used in conjunction with other control methods and is based on the relevant adaptive control
(Wai et al., 2013; Tian et al., 2010; Deng and Gao, 2011; Mian and Wang, 2008; Li and Hu, 2011) .
Backstepping control has the flexibility to avoid the elimination of useful nonlinearity and to
achieve stability and tracking. Therefore, the control method can be easily extended to adaptive
control and sliding mode control without limiting the uncertainty of the matching conditions (Song
and Sun, 2014; Zhou et al., 2011; Vijay and Jena, 2017; Madani and Benallegue, 2007) . In (Vijay
and Jena, 2017), a new type of autonomous model is studied, which reduces the state variable of
the model. The simulation and experiment shows the vality of the proposed controller. In (Ginoya
et al., 2015), a multi-plane sliding mode control method based on disturbance observer is proposed.
The estimation of uncertainty and derivative of virtual input is presented to solve the
non-matching uncertainty and disturbance nonlinear system, the asymptotic stability of the method
is verified, compared with other control strategies, and experimental research is carried out on the
serial robot. In(Adhikary and Mahanta, 2013), combining backstepping control and sliding mode
control, which is related to matching and non-matching uncertainties. A new controller was
proposed and the simulation research is carried out to verify the robustness of the proposed
controller.
In this paper, considering the uncertainty of the robot in the trajectory tracking, a new
controller was proposed. ESO is used to estimate the external disturbance、position and speed. The
new controller is designed, which combines the sliding mode and the backstepping control
strategy. This paper is divided into 4 parts, first of all, on the basis of considering the external
disturbance, according to the dynamics parameter equation of the robot, the related derivation is
given, the related characteristics are introduced. In section 2, this paper designs a controller based
on ESO, fast terminal sliding mode variable structure control and backstepping strategy, The
Barbalat theorem and Lyapunov theorem are used to prove its stability. In section 3, the validity of
the proposed controller is verified by MATLAB simulation experiment and physical experiment,
and the relevant data curve is given, and the practicability of the controller is verified. In section 4,
the experimental is carried out and results are discussed in depth. The results are discussed and the
robustness of the proposed controller are error analysis with the related methods.

2 Kinematics and Dynamics model of Robotic Manipulator

2.1 Robot Kinematics Model
When the robot performs the work task, its controller plans the position and pose according to
the trajectory instruction. Using the inverse kinematics algorithm to calculate the joint parameter
sequence, and then drives the robot joint to make the manipulator move according to the
predetermined position and pose according to the machining trajectory instruction.
Trajectory of robot manipulator:

x  h(q)

The x  R n is the position vector in the robot task space, q  R is the joint angle in the joint
n

space, and h  q   R describes the transition relationship between the joint space and the task
n

space.
2.2 Robot Dynamics Model
For a rigid n-dof manipulator, its dynamic performance can be described by the following
second-order equations:

M ( q) q C( q, q) q G( 
q) F( q) d   (1)

Where, q  R is the angular displacement of joints, M  q   R mn is the inertial matrix

n

for the robot, C  q, q   R n Represents centrifugal force and gothic forces. G  q   R n Represents

gravity items. F  q   R n indicates the frictional moment,   R n indicates Control torque,

 d  R n is the external interference.

The characteristics of the robot dynamic system are as follows:
(1) M (q)  2C (q, q) is a skew symmetric matrix, including x ( M (q)  2C (q, q)) x  0 ;
T
 (2) M  q  satisfies the following bound and m1 , m2 are positive numbers:

m1 ‖ x ‖ 2  xT M (q ) x  m2 ‖ x ‖ 2

(3) M (q) , C (q, q) , G(q) , F (q) satisfy the following relationships

M (q)  C (q, q)   G(q)  F (q)   (q, q,  , ) P (2)

Which  (q, q,  , )  R nn regression function for the known joint variable function.It is
a known function matrix of the robot's generalized coordinates and its derivatives, and P  R is
n

the positional constant parameter vector describing the mass characteristics of the robot.
3 Adaptive ESO Design
According to (Cui et al., 2017) and (Ginoya et al., 2015) ,and equation (1),this paper designs
an adaptive MIMO-ESO.
 x1  x2 (3)

 x2  M ( y )    d  F  x2   G ( y )  C  x1 , x2  
1


 y  x1

x1   q1 , q2 qn  (4)

Assumption1. Assume that all external disturbances  d are set to be continuous and are

bounded. Where y is the output x1 . The specific representation of the extended State

observer is as follows:

 z1  z2  1e1 (5)

 z2  z3   2 Fal1  x2
 z3   3 Fal2

n*1
Which e1  R
n1
represent an observation error vector, zi  R is the state vector

xi  R n1vector of observed values, where i  1, 2,3 , 1 ,  2 , 3 are the gain matrix. Fal1 and

Fal2 is a two different vector matrix (Cui et al., 2017). where,

Fal1   fal11  e11 ,  1 ,   fal12  e11 ,  1 ,    (6)

T

And

Fal2   fal21  e11 ,  2 ,   fal22  e12 ,  2 ,   

T
(7)

The convergence of ESO is obtained from the error system, which is reconstructed from ESO
(3) and system (5). The system error can be described as follows:
 e11  e21  11e11 (8)

 e21  e31   21 fal11
e  y   fal  e 
 31 31 21 11

Theorem 1 Considering the kinematic error (8), Under the Assumption 1 effect, If the
parameter 11 ,In ESO is large enough, Then the estimated error, e11 e21 and e31 is to converge

to zero within a finite time.

Proof1. The stability of ESO is proven as follows:
1 1 1 1 1 1 2 (9)
Veso  1  e11  e21   2  e11  e31   3  e21  e31   4 e112  5e21
2 2 2 2
 6 e31
2 2 2 2 2 2
Which, eij represents the observation error，which i  1, 2,3 ， j  1, 2,3 ，

1 , 2 , 3 non-negative constants, 4 , 5 , 6 are the normal number，To simplify the expression of

the formula,Veso can be expressed as follows:

1 1 1
Veso   1  2  4  e112   1  3  5  e212   2  3  6  e312  1e11e21  2 e11e31
2 2 2
3e21e31 (10)

To simplify the expression of formulas, We do the following variable substitution:

1  1  2  4

2  1  3  5

3  2  3  6
This gives you the following expression:

1 1 1 (11)
Veso  1e112  2 e212  3e312  1e11e21  2 e11e31  3e21e31
2 2 2
Assume 1 , 2 , 3 can be zero, Then the formula can be expressed as follows:

Veso   1 11e112  1e11e21  2  21e21 fal11  e11   3 31e31 fal11  e11   2 e21e31

3e31 y (12)

When considering the real problems and influence factors in the display, the first order differential
of the nonlinear term y is bounded. also exists, Veso  0 If the parameter 11 is chosen to be
large enough. In this case, the trajectory of the error converges to zero. The stability of ESO has
been verified.
4 Design of Robust Sliding Mode Controller With Backstepping
According to the proof 1 and the remark 1, the tracking error can be expressed as:

ei  qdi  qi , ei  qdi  qi (15)

 d  (qi , qi )  dˆ (16)

Where, qdi is the desire i joint displacement , qi is the real i joint displacement, ei is the
error in i joint, d is the drive torque，(q
i , qi ) is the error torque due to the joint，d̂ is the

torque as to the dynamic uncertainty.

Assuming that dynamic uncertainties and changes in external disturbances are very slow, then it
is known:

d  0 (17)

Suppose Yd 、Yd donate the desired trajectory and speed. The proposed controller is

conducted as follows:
STEP1:
Define the trajectory error z1 as follows, according to the manipulator trajectory trackingYd :

z1  Y  Yd (18)

Then

z1  Y  Yd  x2  Yd (19)

The stability factor of the manipulator can be selected as:

1  c1 z1 (20)

Wherein, 1 is the stability coefficient, c1 is the normal number. The speed error of the robot

control system can be expressed as:

z2  z1  1  x2  Yd  1 (21)

The Lyapunov function can be represented as follows:

1 T (22)
V1  z1 z1
2

Then,

V1  zT1 z 2 zT1c z1 1
(23)
If the equation satisfied, z2  0 then the first part of the robot control system is stable.

STEP2
Because

z2  x2  Yd  1 (24)

Re-select the Lyapunov function as follows:

1 (25)
V2  V1  sT s
2
where s is the terminal sliding mode conversion function:

s  k z1  z2 (26)

where, k  0 then
V2  z1T z1  z1T c1 z1  sT   k  z2  c1 z1   M 1 (q )  z2  Yd  1   G (q )   d 
Yd  1 ] (27)

When V2  0 , the control law can be defined as follows:

  Mˆ  k  z2  c1 z1   Yd  1  h(s   sgn (s ))   C (q, q )  z2  Yd  1   G (q )

 F sgn ( s) (28)

STEP 3
A third Lyapunov function is defined as follows:

1 (29)
V3  V2   d dT
2

Which,  is the normal number;  d is an indeterminate kinematic parameter estimation of.

 d Estimated error d   d  ˆd can be descriped as:

 d   d  ˆd   d (30)

Also because of  d  0 ,then

1

 
V3  zT1 z1  zT1 c1 z1  dT ˆ d   M 1 ( q) s  sT  k 2z 1c 1z d Y 1 

  c ( q, q )  z 2  Yd  1   G ( q )   d 
(31)
 M 1 ( q )  
The adaptive law can be chosen as:

ˆ   M 1 (q ) s (32)

The adaptive terminal sliding mode backstepping controller can be designed as follows:

  M  k  z2  c1 z1   Yd  1  h(s   sgn (s ))   C (q, q )  z2  Yd  1  

G(q)  ˆ sgn ( s) (33)

Bring the formula (33) and (32) into the (31) and it can be obtained :

V3  z1T z2 - z1T c1 z1 - hsT s - hbsT ( sgn (s)) (34)

Select the transfer matrix as follows:

 2 1
c1  hk hk  
2
(35)
R 
 hk  1 h 
 2 
Because

zT Rz  c1 z1T z1  z1T z2  hsT s (36)

And

zT Rz  c1 z1T z1  z1T z2  hsT s (37)

Then
1 (38)
‖ R ‖  h  c1  k  
4
By selecting the appropriate h , c1 and k to satisfy R  0 , Transfer matrix R is considered to

be a positive definite matrix, then

V3   zT Rz  h sT ( sgn (s))  0 (39)

Where , z  z1 z2 . Then, the third part of the robot is also stable. The controller block

diagram is proposed as Fig 1 shows in Fig1 shows:

 Backstepping
controller

Yd qd 
Inverse
kinematic   Forward p
FTSMC
kinamatic

ei
. ..
qi qi qi
ESO Force
sensor

Fig. 1. The proposed controller block diagram

5 Simulation And Experiment:
5.1 Simulation
In order to verify the performance of the controller, the simulation experiment uses
matlab/robotic toolbox as the simulation platform, the simulation experiment uses the Bernoulli
lemniscate as the target curve, the 5-DOF connecting rod robot as the experimental carrier, the
terminal actuator position and posture coordinates are  x y z  . The desired curve is as

followed and the unite is（dm）:

 1  0.95cos( t ) / 1  sin 2 ( t )   (40)
 
xd (t )  1  0.95sin( t ) cos( t ) / 1  sin 2 ( t )  
 
 0 
 
During the simulation process, the system model is applied to the parameters uncertainty in
the trajectory tracking. Suppose Mˆ  0.92* M , Cˆ  0.92* C , Gˆ  0.92* G .The D-H and
dynamic parameters are shown in Table 1 and Table 2.
SMC: Sliding mode control is a relatively simple control mode, η=1.2 is taken during the control
law of the η sgns for the robust term. The switching function will cause the control input signal to
produce chattering phenomenon and it will cause the control input layer signal to produce
chattering, and thus affect the control accuracy. The switching function used in this paper can be
replaced by the saturation function. In addition to the boundary layer, the system state quickly
tends to slide mode, and in the boundary layer, the feedback control is used to reduce the
chattering caused by fast switching of the sliding mode.
FTSMC: The fast terminal sliding mode control is a kind of sliding mode control strategy which
is compared with the ordinary sliding mode control. The fast terminal sliding mode control is to
introduce the nonlinear function in the design of the sliding mode surface, and to construct the
terminal sliding mode surface. The tracking error on the sliding surface can converge to zero
within the specified finite time t . In this paper, the saturation function is substituted for the
switching function, the boundary layer thickness is δ=0.2. The nonlinear sliding surface is
3 5
designed as Stsm  x2  x1 5 and Sntsm  x1  x1 3 , the control law parameters are taken

D = 0.3 ,  = 1.0 ,  = 0.023 .The boundary layer thickness and the sliding mode surface
parameters can be found during the simulation, which has an influence on the whole control
process.
The proposed controller: As mentioned above, this controller is a controller combining the
extended state observer, the fast terminal sliding mode control and the backstepping control
strategy, the proposed control gain of the controller used by the robot for the controller is as
follows: c1  2.5 , k  0.6 , h  2.5 ,   2.136 ,   2.136 .The extended state observer is used

to deal with the nonlinear external disturbance of the robot, and in this state observer, not only the
state of the system is estimated, but also the values of all external disturbances need to be
estimated. The basic idea of backstepping control is to decompose a complex nonlinear system
into a subsystem that does not exceed the system.
The simulation results are shown in the following diagram. The robot mechanism parameters、
dynamic parameters in table 1 and table2.
Table 1 Robot Mechanism Parameters

link ai (mm) i (deg ) di (mm) i (deg )

#1 0 90 0 1
#2 0 -90 -145.8 2
#3 1 -90 520.6 0
#4 0 0 -128.9 4
#5 0 90 0 5
Table 2 Robot Dynamics Parameters
Motor Viscous friction
mi  kg  li  m 
i

Inertia Jm, kg  m 2  
coefficients B, N  m  s 1 
L1 5.7 1.2 360e4 1.48e3
L2 3.5 0.75 210e4 1.817e3
L3 1.4 0.75 70e4 7.12e4
L4 1.4 0.46 70e4 6.32e4
L5 0.6 0.15 33e 4 1.16e4
The input torques are as Fig2 shows:

90
SMC

60
JOINT1 (Nm)

30

-30

-60

0.0 2 4 6 8 10 12 14
time (sec)
 90
80
70 ESO-FTSMC
60 FTSMC
50
40
30
JOINT1 (Nm)

20
10
0.0
-10
-20
-30
-40
-50
-60
-70
-80
0 2 4 6 8 10 12
time (s)

70
SMC
JOINT2 (Nm)

-70

-140

0 5 10 15
time (sec)

20

15

10

0
Joint2 (Nm)

-10

-15

-20

-25

-30
FTSMC
-35
ESO-FTSMC
-40

0.0 2 4 6 8 10 12 14
time (s)

50

SMC
25

0
JOINT3 (Nm)

-25

-50

-75

-100

0.0 2 4 6 8 10 12
time (sec)
 15

10 FTSMC
ESO-FTSMC
5

0
JOINT3 (Nm)

-5

-10

-15

-20

-25

-30

0.0 2 4 6 8 10 12 14
time (s)

30

SMC
20
JOINT4 (Nm)

10

-10

-20
0 2 4 6 8 10 12
time (sec)

20
ESO-FTSMC
FTSMC
15
JIONT4 (Nm)

10

-5

-10

0 2 4 6 8 10 12
time (s)

6
SMC

4
JIONT5 (Nm)

-2

-4

0 2 4 6 8 10 12
time (sec)
 FTSMC
4 ESO-FTSMC

2
JIONT5 (Nm)

-2

0 4 8 12
time [s]

Fig2.The input torque of different controller

After the 12 seconds of the entire process. The main characters of the proposed controller are
just as follows:
1 During the simulation, the SMC method appears chattering phenomenon. Considering the
external disturbances, position and speed disturbance, the quality of the controller still need
improve in the process of the sliding surface stability.
2 ITSMC and ESO-SMC can well reduce "chattering" phenomenon.The ESO-SMC can observe
the external disturbance, position and speed, which can be better to apply the external disturbance
to the entire control strategy.
3 The dynamic parameters during the simulation and bound of the uncertainty are also a
influencing factor. Especially the interaction of the sliding mode surface during the process, there
are always unknown factors which we should focus more attention on.
5.2 Experiment
E x t
P C
NDI 3D
Measurement System

c o n t

C a n
Welding
robot X
s e D r A v Q
d r c i a v r
Y
Z
X s I e / r O
m P o o t r
Y
Jr3 force
Z sensor
J o i
l i m
Mainpulator
w e l d
m a 6 i - p a
s e n

(a) (b)
Fig.3.The experiment process equipment (a). NDI and the robot manipulator system (b).The hardware
of the robot manipulator
 z 1
z
y
2
z 0 1 y 2
Observer-based control

Unknown gravitational torque

y x 1 x Eso Unknown friction torque

trajectory
2

Desired
0 Unknown dynamics
Robot
x
Estimate of the
0  2
unknown torque

 1
d 1 d 2
d 3
Controller Dynamics

z 3
 3
Estimate of
the position
Position and
velocity
Position

x 3 y 3
Manipulator
Eso Kinematics
 4

Uncertain kinematics

(c) (d)
Fig.4. The structure of the robotic and controller (a). Schematic diagram of robot mechanism (b). Block
diagram of the proposed controller
Figure 3 and 4 are the welding robot, six-D force sensor, NDI laser measuring equipment, etc
which will be used to perform the proposed controller in the process of experiment.
The mark point is established at the robot manipulator, which connects the NDI device with the
robot terminal actuator and a world coordinate system is formed. During the experimental process,
we also need to consider the external disturbances and internal disturbances, so the devices in the
environmental are more stringent. In the course of the experiment, the manipulator runs at 2m/s
speed, and the six-D force sensor of the manipulator of the welding test is 2.2 lb to 1000b, and the
force sensing accuracy is 1 to 50N.

A: The first point B: The second point

C: The third point D: The forth point

 E: The fifth point F: The terminal point
Fig5.The trajectory tracking diagram with the proposed controller

10

0
Torque(mn.m)

-10

-20

-30
Disturbance
Estimated Value
-40
0 1 2 3 4 5
time (s)

Fig. 6 State diagram of external disturbance estimation

The entire experiment was carried out around the manipulator along the trajectory, as Fig5
shows. The total time was 5 seconds. The whole experiment is divided into 3 parts：firstly，the
robot reached the target origin point, and then move along the set trajectory. we can increase the
excitation in the course of the motion, the increase of the excitation is U=5[1-cos (6.28t)][1-exp
(-t)]. Fig.6 shows the values. We set O, A, B, C, D, E points throughout the trajectory tracking
process to facilitate the observation of the progress of the whole experiment and related research.
Testing of the different controllers is not less than 5 times, the test data is recorded, and related
analysis, in order to accurately compare the tracking error.
During the test the process, we can increase the thickness of the boundary layer, so as to
facilitate the control of buffeting phenomenon on the whole experiment.
 s s (39)
 , if 1
s   
sat    
  s
sgn   , otherwise
 
The whole process of the experiment result about force and errors were shown as Fig7and Fig8
show.
 1.4 x

1.0

y
-1.12
Force [N]

-1.16
-1.12

z
-2.62

-2.80

0 1 2 3 4 5
time [s]

(a)ESO-FTSMC experiments

1.4 x

1.2

1.0

y
Force [N]

-1.52
-1.53
-1.54

-1.51 z

-1.52

0 1 2 3 4 5
time [s]

(b) PID experiment

Fig7. The external forces measured in the X, Y, Z direction of the wrist Force sensor (FS)
x
-0.4185

-0.4200

-0.4215
0 1 2 3 4 5
y
-0.08
Errors [mm]

-0.09

0 1 2 3 4 5
0.2 z
0.0
-0.2
-0.4
0 1 2 3 4 5
time [s]

a. the ESO-FTSMC experiment.

 -1.53 x

-0.756 0 1 2 3 4 5
Error [mm]

y
-0.760

0 1 2 3 4 5
1.4
z
1.3

0 1 2 3 4 5
time [s]

b . the PID experiment.

Fig.8.An error distribution of the manipulator in the Descartes coordinate system

Table3 The RAEs of the joint torque estimations based on parametric adaptions
tests Joint1 Joint2 Joint3 Joint4 Joint5
1 0.0225 0.0256 0.0341 0.0254 0.0524
2 0.0165 0.0247 0.0214 0.0247 0.0571
3 0.0144 0.0254 0.0328 0.0218 0.0624
4 0.0158 0.0147 0.0361 0.0301 0.0637
5 0.0224 0.0354 0.0357 0.0314 0.0601
6 0.0234 0.0247 0.0351 0.0341 0.0587
means 0.0192 0.0251 0.0325 0.0279 0.0591

In order to verify the effectiveness of the above controller, the relevant error analysis method
is described in Table 3, the RAEs of the joint torque can be maintained within [0.019, 0.06], which
estimates the relative accuracy of the joint torque and the robot parameters. In(Yin and Li, 2018),
the quality of the proposed controller can be measured and estimated. The trajectory tracking error
is analysis in Fig9.
1.0
Averaged relative tracking error(%)

0.5

0.0

-0.5

-1.0

-1.5

-2.0 The PID

control
·· The proposed
control
-2.5
0 1 2 3 4 5
Time
(a) Averaged relative tracking error(%)
 60
The PID Control

Normalized averaged control siginal (%)

The proposed
control
40

20

-20

-40

-60
0 2
1 Time 3 4 5

(b) Normalized averaged control signal(%)

Fig9. Schematic diagram of robot trajectory tracking error analysis
6 Conclusion：
Aiming at the problem of disturbance, uncertain dynamics and unknown torque in the course
of robot trajectory tracking, an extended state observer was designed, and the new controller was
proposed. The main contributions of this article were as follows:
1 This paper combines the ESO (Extend state observer), sliding mode control and
backstepping method to enhance the robotic dynamic robustness. Backstepping method reduced
the chattering phenomenon during the dynamic controller. Considering the dynamic uncertainty
and kinematic uncertainty, the new controller was proposed to handle the problems in the process
of the trajectory tracking.
2 The new controller was conducted on the micro-macro welding robot, and finally the results
showed that the proposed controller can guarantee the trajectory tracking error to be zero. ESO
was used to observe the dynamic uncertainty, joint position and velocity in the process of
trajectory tracking, and finally realized input torque to be optimized.
3 The new experimental device and modern test equipment were introduced. Micro-macro
robot, NDI and six-dimensional force sensor were used to performance the result of the proposed
method. And the results showed that the tracking accuracy was realized within 0.3mm considering
the dynamic uncertainty and disturbance.

7 Acknowledgments

This research was supported by National Key Research and Development Program of China
(2018YFB1308301) and National Natural Science Foundation of China (No.61374133,
No.61673205). Data provided by Nanjing university of science and technology robotics
institute.
8 Reference：
Adhikary N and Mahanta C. (2013) Integral backstepping sliding mode control for underactuated
systems: Swing-up and stabilization of the Cart–Pendulum System. Isa Trans 52: 870-880.
Capisani LM and Ferrara A. (2011) Trajectory planning and second-order sliding mode
motion/interaction control for robot manipulators in unknown environments. Ieee
Transactions on Industrial Electronics 59: 3189-3198.
Chan L, Naghdy F and Stirling D. (2013) Extended active observer for force estimation and disturbance
rejection of robotic manipulators. Robotics & Autonomous Systems 61: 1277-1287.
Chen JY. (2001) Rule regulation of fuzzy sliding mode controller design: direct adaptive approach.
Fuzzy Sets & Systems 120: 159-168.
Cui R, Chen L, Yang C, et al. (2017) Extended State Observer-Based Integral Sliding Mode Control for
an Underwater Robot with Unknown Disturbances and Uncertain Nonlinearities. Ieee
Transactions on Industrial Electronics 64: 6785-6795.
Deng L and Gao S. (2011) The design for the controller of the linear inverted pendulum based on
backstepping. International Conference on Electronic & Mechanical Engineering &
Information Technology.
Fallaha CJ, Saad M, Kanaan HY, et al. (2011) Sliding-Mode Robot Control With Exponential Reaching
Law. Ieee Transactions on Industrial Electronics 58: 600-610.
Ferrara A and Incremona GP. (2015) Design of an Integral Suboptimal Second-Order Sliding Mode
Controller for the Robust Motion Control of Robot Manipulators. Ieee Transactions on
Control Systems Technology 23: 2316-2325.
Ginoya D, Shendge PD and Phadke SB. (2015) Disturbance observer based sliding mode control of
nonlinear mismatched uncertain systems. Communications in Nonlinear Science & Numerical
Simulation 26: 98-107.
Guo Y and Woo PY. (2003) An adaptive fuzzy sliding mode controller for robotic manipulators.
Systems Man & Cybernetics Part A Systems & Humans IEEE Transactions on 33: 149-159.
Li HY and Hu YA. (2011) Robust sliding-mode backstepping design for synchronization control of
cross-strict feedback hyperchaotic systems with unmatched uncertainties. Communications in
Nonlinear Science & Numerical Simulation 16: 3904-3913.
Lu B, Fang Y and Sun N(2017). Sliding mode control for underactuated overhead cranes suffering from
both matched and unmatched disturbances. Mechatronics.47:16-125.
Madani T and Benallegue A. (2007) Sliding Mode Observer and Backstepping Control for a Quadrotor
Unmanned Aerial Vehicles. American Control Conference.5887-5892.
Mian AA and Wang D. (2008) Modeling and Backstepping-based Nonlinear Control Strategy for a 6
DOF Quadrotor Helicopter. Chinese Journal of Aeronautics 21: 261-268.
Park KB and Lee JJ. (1997) Sliding mode controller with filtered signal for robot manipulators using
virtual plant/controller. Mechatronics 7: 277-286.
Precup RE, Radac MB, Roman RC, et al. (2017) Model-free sliding mode control of nonlinear systems:
Algorithms and experiments. Information Sciences 381: 176-192.
Qiao L and Zhang W. (2017) Adaptive nonsingular integral terminal sliding mode tracking control for
autonomous underwater vehicles. Iet Control Theory & Applications 11: 1293-1306.
Qiao L and Zhang W. (2019) Double-Loop Integral Terminal Sliding Mode Tracking Control for UUVs
 With Adaptive Dynamic Compensation of Uncertainties and Disturbances. IEEE Journal of
Oceanic Engineering .44(1):29-53.
Song Z and Sun K. (2014) Adaptive backstepping sliding mode control with fuzzy monitoring strategy
for a kind of mechanical system. Isa Transactions 53: 125-133.
Tang Y. (1998) Terminal sliding mode control for rigid robots ☆. Automatica 34: 51-56.
Tian Z, Wu H and Feng C. (2010) Hierarchical adaptive backstepping sliding mode control for
underactuated space robot. International Asia Conference on Informatics in Control.
Tran MD and Kang HJ. (2016) Adaptive Terminal Sliding Mode Control of Uncertain Robotic
Manipulators Based on Local Approximation of a Dynamic System. Neurocomputing 228:
S0925231216312759.
Vijay M and Jena D. (2017) Backstepping terminal sliding mode control of robot manipulator using
radial basis functional neural networks. Computers & Electrical Engineering 67:
S0045790617312004.
Wai RJ, Lin CY, Wu WC, et al. (2013) Design of backstepping control for high-performance inverter
with stand-alone and grid-connected power-supply modes. Iet Power Electronics 6: 752-762.
Yagiz N and Hacioglu Y. (2009) Robust control of a spatial robot using fuzzy sliding modes.
Mathematical & Computer Modelling 49: 114-127.
Yin X and Li P. (2018) Enhancing trajectory tracking accuracy for industrial robot with robust adaptive
control. Robotics and Computer-Integrated Manufacturing 51: 97-102.
Zhao D, Zhu Q and Dubbeldam J. (2015) Terminal sliding mode control for continuous stirred tank
reactor. Chemical Engineering Research & Design 94: 266-274.
Zhou, Jianda and Xianzhong. (2011) Backstepping Based Global Exponential Stabilization of a
Tracked Mobile Robot with Slipping Perturbation. Journal of Bionic Engineering 8: 69-76.