Automatica 64 (2016) 7075

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Brief paper

Barrier Lyapunov Functions-based adaptive control for a class of

nonlinear pure-feedback systems with full state constraints
Yan-Jun Liu 1 , Shaocheng Tong
College of Science, Liaoning University of Technology, Jinzhou, Liaoning, 121001, China

article

info

Article history:
Received 13 January 2015
Received in revised form
21 July 2015
Accepted 2 October 2015
Available online 7 December 2015
Keywords:
Nonlinear pure-feedback systems
Adaptive control
Full state constraints
Barrier Lyapunov Functions

abstract
In this study, an adaptive control technique is developed for a class of uncertain nonlinear parametric
systems. The considered systems can be viewed as a class of nonlinear pure-feedback systems and the
full state constraints are strictly required in the systems. One remarkable advantage is that only less
adjustable parameters are used in the design. This advantage is first to take into account the pure-feedback
systems with the full state constraints. The characteristics of the considered systems will lead to a difficult
task for designing a stable controller. To this end, the mean value theorem is employed to transform the
pure-feedback systems to a strict-feedback structure but non-affine terms still exist. For the transformed
systems, a novel recursive design procedure is constructed to remove the difficulties for avoiding nonaffine terms and guarantee that the full state constraints are not violated by introducing Barrier Lyapunov
Function (BLF) with the error variables. Moreover, it is proved that all the signals in the closed-loop
system are global uniformly bounded and the tracking error is remained in a bounded compact set. Two
simulation studies are worked out to show the effectiveness of the proposed approach.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
In real world, the practical plants are inevitable to contain the
uncertainties. Driven by practical requirements and theoretical
challenges, the controller design of uncertain systems has become
an important research domain (Li, Gao, Shi, & Zhao, 2014; Li, Pan, &
Zhou, 2015; Li, Yin, Pan, & Lam, 2015). Adaptive control algorithms
on nonlinear systems with uncertain parameters have received
much attention (Krstic, Kanellakopoulos, & Kokotovic, 1995; Krstic
& Kokotovic, 1996; Wang & Lin, 2010, 2012; Yang, Ge, & Lee,
2009; Zhang & Lin, 2015; Zhang & Xu, 2015). Subsequently, some
significant works on adaptive NN control for nonlinear systems
with unknown functions were revealed in Chen, Hua, and Ge
(2014), Ge and Wang (2004), Liu and Tong (2015) and Tong, Sui,
and Li (2015); Tong, Zhang, and Li (in press). Other notable adaptive
control methods were proposed in He, Ge, How, Choo, and Hong
(2011); He, Zhang, and Ge (2014a,b) and Li, Xiao, Yang, and Zhao

This work was supported by the National Natural Science Foundation of China
under Grants 61374113 and 61473139, and Program for Liaoning Excellent Talents
in University (LR2014016). The material in this paper was not presented at any
conference. This paper was recommended for publication in revised form by
Associate Editor Shuzhi Sam Ge under the direction of Editor Miroslav Krstic.
E-mail addresses: liuyanjun@live.com (Y.-J. Liu), jztongsc@163.com (S. Tong).
1 Tel.: +86 4164198736; fax: +86 4164198736.

http://dx.doi.org/10.1016/j.automatica.2015.10.034
0005-1098/ 2015 Elsevier Ltd. All rights reserved.

(2015) for some real systems with uncertain parameters. Adaptive

control for a more general class of pure-feedback systems with the
uncertainties was studied in Ge and Wang (2002), Ge, Yang, and
Lee (2008), Krstic et al. (1995) and Nam and Araposthathis (1988).
However, the effect of the constraints is omitted in the abovementioned results.
It is a fact that many real systems suffer from the effect
of the constraints, such as the temperature of chemical reactor
and physical stoppages. Two control approaches were proposed
in Zhou, Duan, and Lin (2011); Zhou, Li, and Lin (2013) for
linear discrete-time systems with input constraints subject to
actuator saturation. In Chen, Ge, and Ren (2011), an adaptive
tracking control was constructed for uncertain multi-input and
multi-output nonlinear systems subject to non-symmetric input
constraints. The effect of input constraints is overcome by
introducing an auxiliary design system. But these approaches
cannot solve the control problem of the state constraints.
Recently, the utilization of BLF for solving the control problem
of nonlinear systems with output and state constraints has been an
active area. A survey paper on constraint-handling ways of model
predictive control (Mayne, Rawlings, Rao, & Scokaert, 2000) was
well established. It has also employed BLFs to design the adaptive
controller for nonlinear strict-feedback systems with the timeunvarying (Tee, Ge, & Tay, 2009a) and time-varying (Tee, Ren,
& Ge, 2011) output constraints. In Ren, Ge, Tee, and Lee (2010),

Y.-J. Liu, S. Tong / Automatica 64 (2016) 7075

an adaptive neural control was presented for a class of nonlinear

systems with unknown functions, immeasurable states and the
output constraint. A BLF is effectively used in the output-feedback
control. Moreover, the control problem of some practical systemsbased BLF has been coped with, such as a flexible crane system
(He et al., 2014a,b). Besides the output constraint mentioned in
the above works, the state constraints have also been tackled
by using BLF. Adaptive control techniques-based on BFLs have
been addressed for nonlinear strict-feedback with partial state
constraints (Tee & Ge, 2011) and the full state constraints (Liu, Li,
& Tong, 2014; Tang, 2014; Tee & Ge, 2012; Tee, Ge, & Tay, 2009b).
These techniques on state constraints can be only to control a
class simple of nonlinear strict-feedback systems. To guarantee
the stability of more general class of systems, an adaptive neural
control approach was studied in Kim and Yoo (2014) for nonlinear
pure-feedback systems with full state constraints. However, the
adaptive design of this approach is obtained based on the neural
weight vector. When the neural network nodes are increased, the
number of adjustable parameters will be enormous. Then, the online learning time becomes very large.
In this paper, an adaptive control technique is studied for a class
of nonlinear parametric systems. The main contributions of the
proposed approach are that:
(1) Different from the results in Liu et al. (2014), Tee and Ge (2011)
and Tee et al. (2009b) which focused on the strict-feedback
systems with state constraints, this paper frames a generalization of the results for a more general class of nonlinear purefeedback systems with the full state constraints. At the same
time, in contrast to the approach in Kim and Yoo (2014), less
adjustable parameters are used in the design. It is first to design
an adaptive control technique with less adjustable parameters
to treat the full state constraints of nonlinear pure-feedback
systems.
(2) In order to control this class of systems, the pure-feedback
systems are transformed into a strict-feedback structure with
non-affine terms by using the mean value theorem. For the
transformed systems, a modified backstepping design based on
BLFs is proposed to prevent the violation of the full state constraints and the adaptation laws for the estimations of norms
on uncertain parameters are constructed.
Finally, it is proved that all the signals in the closed-loop system
are global uniformly bounded and the tracking error is remained in
a compact set. The effectiveness of the approach can be shown by
working out two simulation examples.
2. System description and preliminaries
Consider the nonlinear pure-feedback systems (Ge & Wang,
2002)
x i = fi (x1 , . . . , xi+1 ) ,
x n = fn (x, u)
y = x1

i = 1, . . . , n 1

(1)

where x0i+1 is some point between zero and xi+1 , and u0 is some
point between zero and u.
In this paper, the nonlinear smooth functions fi (xi+1 ) , i =
1, . . . , n 1 and fn (x, u) are uncertain, and they satisfy the
following condition
fi (xi+1 ) = iT i (xi+1 ) ,

i = 1, . . . , n 1

fn (x, u) = n (x, u)
T
n

(4)
(5)

where i Rm is an uncertain constant vector; i (xi+1 ) , i =

1, . . . , n 1 and n (x, u) are known continuous function vectors.
The design objective of this study is to construct an adaptive
state feedback controller u such that the tracking error z1 = y
yd (t ) converges to a bounded compact set, all the signals in the
closed-loop system are global uniformly bounded and the full state
constraints are not violated where yd (t ) R is the reference signal
to be known and bounded.
To achieve the control objective, it is necessary to make the
following assumptions.
Assumption 1 (Ge & Wang, 2002). The functions i (), i =
1, . . . , n are bounded, i.e., there exist the constants i > 0
i
such that |i ()| i . Without losing generality, this paper
i
assumes that i () i .
i

Assumption 2 (Tee & Ge, 2011). It is assumed that yd (t ) and its ith
(i)
order
yd (t ) , i = 1, . . . , n satisfy |yd (t )| A0 < kc1
derivatives

(i)

and yd (t ) Ai where A0 , A1 , . . . , An are positive constants.

Remark 1. In Tee and Ge (2011, 2012) and Tee et al. (2009a,

2011), adaptive output and state constraint control problems were
addressed for uncertain nonlinear strict-feedback systems, i.e.,
fi (.) is a function of xi . From (1), it can be seen that fi (.) is not only
a function of xi , but also a function of xi+1 with u = xn+1 . Thus, the
system structure is more general than the previous results.
3. The controller design and stability analysis
Define the tracking error z1 = y yd (t ) and the variables
zi = xi i1 , i = 2, . . . , n where i1 is a virtual controller to
be designed in Step i. The detailed backstepping design is given as
follows.
Step 1: The time derivative of z1 = x1 yd (t ) is
z1 = x 1 y d (t ) = f1 (x1 , x2 ) y d (t ) .

(6)

Using (2) and (4), (6) becomes

z1 = 1T 1 (x1 ) + 1 x1 , x02 x2 y d (t ) .

(7)

Introduce the variable z2 = x2 1 and we have

where x = [x1 , . . . , xn ]T Rn , y R and u R are the states, the

output, the input of the systems, respectively; all the states are constrained in the compact sets, i.e., xi is required to remain in the set
|xi | < kci with kci being a positive constant; let xi = [x1 , . . . , xi ]T
Ri ; fi (xi+1 ) , i = 1, . . . , n 1 and fn (x, u) are nonlinear smooth
functions and continuously differentiable with respect to the argument xi+1 and u, respectively. Let i (xi+1 ) = fi (xi+1 ) / xi+1
and n (x, u) = fn (x, u) / u.
By using the mean value theorem (Ge & Wang, 2002), there
must exist x0i+1 and u0 such that
fi (xi+1 ) = fi (xi , 0) + i

71

xi x0i+1

fn (x, u) = fn (x, 0) + n x, u

xi+1 ,

i = 1, . . . , n 1

(2)
(3)

z1 = 1T 1 (x1 ) + 1 x1 , x02 (z2 + 1 ) y d (t ) .

(8)

The virtual controller 1 is designed as

1 = 1 z1 1 Kz1 1 (x1 ) /212 Kz1 1 /2

(9)

where 1 and 1 are the positive design

1 (x1 ) =
parameters,

1 (x1 )2 , 1 = (yd (t ))2 , Kz1 = z1 / k2b1 z12 , 1 > 0 is the

estimation of 1 = 1 2 , kb1 = kc1 A0 is a positive constant. In

the following, Kzi = zi / k2bi zi2 , i 2 will be used and define a

compact set zi := |zi | < kbi , i = 1, . . . , n where kbi is specified
later on.

72

Y.-J. Liu, S. Tong / Automatica 64 (2016) 7075

Introduce the variable zi+1 = xi+1 i and we have

Define a positive definite Lyapunov function

V1 =

1
2

log

k2b1

k2b1 z12

1 1
2 1

zi = iT i (xi ) i1 + i xi , x0i+1 (zi+1 + i ) .

12

(10)

where log () stands for the natural logarithm of , 1 = 1

1 and 1 is a positive design parameter. In the set z1 , V1 is
continuous. Then, the time derivative of V1 is given by V 1 = Kz1 z1 +

1 1 1 /1 .

The virtual controller i is designed as

i = i zi i Kzi i (xi ) /2i2

zi k2bi zi2 Kz2i1 /2 Kzi i /2

+ Kz1 1 x1 , x2 1 Kz1 y d + 1 1 1 /1 .

(11)

i (xi ) = i (xi )2 +

i1

i1

i =

Kz1 1T 1 (x1 ) 1 Kz21 1 (x1 ) /212 + 12 /2 ,

1

yd

y(dj)

2
2
i1
i 1

i1
i1
+
j +
x j +1 .
xj
j
j =1
j =1

Kz1 1 x1 , x2 z2 Kz21 z22 /2 + 21 /2 ,

2

Kz1 y d 1 Kz21 1 /2 + 1/21 .

It is obvious that Kz1 1 0 and thus, we have

K 2 1
1 z1

/2 1 1 Kz1 z1 .

Vi = Vi1 +

1
2

log

V 1 1 Kz1 z1 + 1
1

1
1

1 12

1 21

1 1
2

Kz21 1

+
2

(x1 )

(12)

where c1 =

1
2

(14)

(15)

2
1 1
2

(16)

Using (2) and (4), (16) becomes

zi = iT i (xi ) + i xi , x0i+1 xi+1 i1

(17)

where

i1 =

i (xi )

j =1

xj

j xj +

i 1

i1
j =1

i i (xi ) /2 + /2i

2
i 1
i1

i1 (j+1) 1
i1 (j+1)
i1
2
Kz i
y
i K z i
y
+
(j) d
(j) d
2
2

y
j=0
j =0
i
d
d
2

1
i

1
i1
i1
1
i1
Kz i
j i Kz2i
j +

2
2
j
j=1 j
j =1
i

Kzi iT

i 1

i1
j =0

y(dj)

K2
i zi

(j+1)

i1

i1
j xj , x0j+1 xj+1 +
j .

xj
j =1 j

2
i

i 1

i1

Kzi i xi , x0i+1 zi+1

yd

2
i

i1

1 2
j xj , x0j+1 xj+1
j
xj
2 j=1
j =1
i
2
i 1

1
i1
2
+ i Kzi
x j +1
2
xj
j =1

K z i

i 1

i1 2
/2 2

x
j
j
i

x
j =1

zi = x i i1 = fi (xi+1 ) i1 .

T
j

jT j xj (i 1) i2 /2i + i Kz2i
i

+ 12 1 .
1
Step i: The time derivative of zi = xi i1 is

i 1

i1

xj

= 1 1 1 = 1 12 + 1 1 1

i1

i1
j =1

2
1 1
2

(24)

Using the Youngs inequality, we have

(13)

1 12 + 2 Kz21 z22 + c1

1 12 +

+ Kzi i xi , x0i+1 i Kzi i1 + i i i /i .

Using (14), it has

V i = V i1 + Kzi iT i (xi ) + Kzi i xi , x0i+1 zi+1

K z i

Substituting (19) into V i leads to

K 2 z2
2 z1 2

1 12 + 1 12 .

(23)

Using (13), we have

V 1 1 Kz1 z1

i2 .

2 i

tive of Vi is given by V i = V i1 + Kzi zi + i i /i .

2
1

1 = 1 1 + 1 Kz21 1 (x1 ) /212 .

Kz2 1 (x1 )
1
1 2
1
21

zi2

In the set zi , k2bi / k2bi zi2 is continuous. Then, the time deriva-

The adaptation law for 1 is designed as

1 i

k2bi
k2bi

Using the above four inequalities, (11) becomes

1 1

(22)

Define a positive definite Lyapunov function

Kz1 1 x1 , x02 1 1 Kz21 1 (x1 ) /212

(j+1)

(21)

j =0

i 1

i1 2

x
j
j
x

j=1

Using the Youngs inequality, we have

(20)

2
estimation of i = max1ji j and

V 1 = Kz1 1T 1 (x1 ) + Kz1 1 x1 , x02 z2

(19)

where i and i are the positive

design parameters, i is the

Substituting (7) into V 1 leads to

1
2

i+1 Kz2i zi2+1 +

1 2i
2

i +1

It is obvious that Kzi i 0 and thus, one has

Kzi i xi , x0i+1 i i Kzi zi i Kzi i (xi ) /2i2

(18)

i Kz2i i1 /2 i zi2 Kz2i1 /2.

Y.-J. Liu, S. Tong / Automatica 64 (2016) 7075

Using (18), (20) and the above inequalities, (24) becomes

Theorem 1. Consider the systems (1), and Assumptions 1 and 2 hold.

On the sets zi , the virtual controllers i , i = 1, . . . , n 1
in (9), (20) and the actual controller u in (32), and the adaptation
laws in (13), (26), (33) are constructed. If the design parameters i , i
and i are appropriately chosen to satisfy

V i V i1 i Kzi zi + Kz2i zi2+1 /2 + (i 1) i2 /2

i
i+1
i
i 1

+ i2 /2i + (i 1) /i +

73

2j /2i + 2i /2i+1

j=1

+ i i i /i Kz2i i (xi ) /2i2 .

(25)

The adaptation law for i is designed as

i = i i + i Kz2i i (xi ) /2i2 .

(26)

i i

i
i

Kz2i i

(xi ) /2

i + i .

2
i

2
i

the proposed adaptive control scheme can guarantee that (1) the
tracking error can converge to a bounded compact set; (2) all the
signals in the closed-loop system are global uniformly bounded,
and (3) the full state constraints are not violated.
Proof. From (10), (23) and (34), we have

Similar to (14) and using (26), we have

i = max i xi , j , yd , y(dj) , j = 1, . . . , i

kci+1 > i + kbi+1 ,

2
i

(27)

Vn =

n
1

2 j=1

k2bj

log

k2bj

zj2

n
1 j

It is a fact that log k2bj / k2bj zj2

Using (27), it has

(28)

where ci = (i 1) /2 + /2 + j=1 /2 + (i 1) /
i
i
i
i + 2i /2i+1 + i /2. In Step i 1, it has been obtained that
V i1

j j Kzj zj

j =1

i1

i1

2
i

i1
1

2 j =1

2
j

2
j

(29)

V i

j j Kzj zj

j =1

i
1

2 j =1

j +
2
j

(30)

Step n: The derivative of zn = xn n1 is

zn = x n n1 = fn (x, u) n1 .

(31)

Construct an actual control input

u = n zn n Kzn n (x) /2n2

zn k2bn zn2 Kz2n1 /2 Kzn n /2.

(32)

Design the adaptation law for n

n = n n + n Kz2n n (x) /2n2 .

(33)

k2bn e2n

j=1

j j Kzj zj

n
1

2 j =1

(37)

Using (37), the inequality (35) can be represented as

V n Vn +

(38)
1

1 n
2 n

n2 .

j j2 +

(34)

cj .

(39)

j = j + j , j must be bounded.
From x1 = z1 + yd (t ) and |yd (t )| A0 , we can obtain |x1 |
|z1 | + |yd | < kb1 + A0 . Let kb1 = kc1 A0 and then, |x1 | < kc1 . It can
be known from the definition of 1 (.) in (9), 1 (.) is a function of
, x1 , z1 and y d . Because the boundedness of 1 , x1 , z1 and y d , 1 (.)
is bounded and satisfies |1 (.)| 1 . Then, |x2 | |1 | + |z2 |
1 + kb2 . This implies that |x2 | < kc2 if kb2 = kc2 1 . Similarly, it
can in turn be proven that |xi+1 | < kci+1 , i = 2, . . . , n 1 as long
as kbi+1 = kci+1 i . From the definition in (32), we can see that u
(n)
is a function of j , x and yd , y d , . . . , yd . Owing to the boundedness
(n)
of j , x and yd , y d , . . . , yd , the controller u is bounded.
From the above analysis, we can conclude that all the signals
of the closed-loop system u, x and are bounded and the system
states are not violated.
From (39), it is easy to obtain

We take
on both sides of above inequality, it has
exponentials

k2b1 / k2b1 z12

Vn (t ) [Vn (0) / ] e t + /

Using (30) with i = n 1 and the similar procedure given in Step

i, we have
V n

j2 .

log k2b1 / k2b1 z12 2 [Vn (0) / ] e t + 2/.

Define the Lyapunov function candidate

k2bn

2 j=1 j

log k2bj / k2bj zj2 and j are bounded. Because j is bounded and

j =1

log

n
1 j

Then, it can be concluded

from the above inequality and (36) that

cj

+ i+1 Kz2i zi2+1 .

2 j =1

Kzj zj +

Vn (0) + /.

Vn = Vn 1 +

n
1

where = min 2j j , j , j = 1, . . . , n and =

j = 1 cj .
both sides by e t , (38) can be rewritten as
Multiplying

d Vn (t ) e t /dt e t and integrating it over [0, t], it has

Then, it has from (29) that

i

Vn

cj + i Kz2i1 zi2 .

j =1

< zj2 / k2bj zj2 in the

+ i+1 Kz2i zi2+1 /2 i Kz2i1 zi2 /2

i1

(36)

interval zj < kbj (Ren et al., 2010). Then, (36) becomes

V i V i1 i Kzi zi + ci i2 /2
i
i

2
i
2
i

j2 .

2 j=1 j

(35)

j =1

The following theorem is given to point out the stability of the

closed-loop system.

t +2/

e2[Vn (0)/ ]e

. It is straightforward to get

t
e2[Vn (0)/ ]e 2/

|z1 | kb1 1
= . If Vn (0) = / ,

then, it holds |z1 | kb1 1 e2/ = . If Vn (0) = / ,

it can be concluded that given any > kb1 1 e2/ , there

exists T such
As t ,
that for any t > T , it has |e1 | .
|e1 | kb1 1 e2/ . This implies that |e1 | kb1 1 e2/ .
We can see that e1 can be made arbitrarily small by selecting the
design parameters appropriately.
This completes the proof.

74

Y.-J. Liu, S. Tong / Automatica 64 (2016) 7075

Fig. 1. (a) The trajectories of x1 and yd and (b) phase portrait of z1 , z2 and z3 .

4. Simulation study
Example 1. Consider the following nonlinear systems

x = x1 + x2 + x32 /5

1
x 2 = 0.2 [x1 x2 + x3 + sin (0.2x3 )]
2 2

x 3 = x1 x3 + x2 + 0.2u
y = x1

(40)

where the states are constrained in |x1 | < 0.8, |x2 | < 2 and

T
|x3 | < 2.5; 1 = [1, 1, 0.2]T , 1 (x) = x1 , x2 , x32 , 2 =
[0.2, 0.2, 0.2]T , 2 (x) = [x1 x2 , x3 , sin (0.2x3 )]T , 3 = [1, 1, 0.2]T ,

T
3 (x, u) = x1 x33 , x2 , u . The initial values of the states are
x1 (0) = 0.5, x2 (0) = 0.1, x3 (0) = 0. In this simulation, the
reference signal is yd (t ) = 0.5 cos (0.5t ).
Based on the design procedure, the controller is designed as
u = 3 z3 3 Kz3 3 (x) /232 z3 k2b3 z32 Kz22 Kz3 3

(41)

where z3 = x3 2 and the virtual controller 2 as

Fig. 2. (a) The trajectories of x1 , x2 and x3 and (b) The trajectories of 1 , 2 and 3 .

where the initial value of the adaptation laws is 1 (0) = 0.1,

2 (0) = 0.1 and 3 (0) = 0.1.
The design parameters are chosen as 1 = 5, 2 = 9, 3 = 9,
1 = 1, 2 = 2, 3 = 6, 1 = 2, 2 = 2, 3 = 5. Similar to Tee and
Ge (2011), we can obtain kb1 = 0.3, kb2 = 1.652 and kb3 = 2.931
by using the Matlab routine.
Fig. 1(a) shows the system output tracking trajectory and it
can be seen from this figure that a good tracking performance is
achieved. Fig. 1(b) is given to explain phase portrait of z1 , z2 and
z3 . From this figure, it can be determined that the bounds for z1 ,
z2 and z3 are not overstepped. The system states are illustrated
in Fig. 2(a) and the full state constraints are also not overstepped.
The boundedness trajectories of the adaptation laws are given in
Fig. 2(b).
Example 2. Consider a single-link robot and its dynamic equations
are
M q +

1
2

mgl sin q = u,

y=q

(42)

z2 = x2 1 ,

where q is the angle, u the input torque, M the moment of inertia, g

the acceleration due to gravity m and l are the mass and the length
of the link. The robot parameters are m = 1, l = 1, M = 0.5 and
g = 9.8. The system (1) can be represented as

1 (x1 ) = 1 (x1 )2 ,

x 1
0

x = 0

2 = 2 z2 2 Kz2 2 (x) /222 z2 k2b2 z22 Kz21 Kz2 2

1 = 1 z1 1 Kz1 1 (x1 ) /212 Kz1 1 /2

with z1 = y yd (t ), Kzj = zj / k2bj zj2 , j = 1, 2, 3

1 = (yd (t ))2 ,

2
1

1 (x1 )
2 (x2 ) = 2 (x2 ) +
,
x1

2 2
,

3 (x3 ) = 3 (x3 )2 +

x
j
j
x

2 =
3 =

1
1
y d
+
y d
yd
y d

2
2

2 (j+1)
2

1
1
1

yd

y(dj)
j =0

2
2
2
2

2
2
+
j +
x j +1 .
xj
j
j =1
j =1

The adaptation laws are designed as

j = j j + j Kz2j j xj /2j2 ,

1
2

x1
0
+
x2
1

mgl sin x1

/M

(43)

y = x1

where x1 = q and x2 = q , and the states are constrained in |x1 | < 1,

|x2 | < 1. The initial value of the states is x1 (0) = 0.5, x2 (0) = 0.1.
In this simulation, the reference signal is yd (t ) = 0.5 cos (t ).

j =1

1
0

j = 1, 2, 3

1
x2
x1

The design parameters are chosen as 1 = 5, 2 = 6, 1 = 4,

2 = 6, 1 = 2, 2 = 3. Similar to Tee and Ge (2011), we can
obtain kb1 = 0.5, kb2 = 1.231 by using the Matlab routine. The
simulation results are obtained in Fig. 3. From this figure, we can
see that the tracking performance is good to be achieved, and the
constraints are not violated.
5. Conclusion
In this study, a novel adaptive control technique has been developed for nonlinear pure-feedback systems with the full state
constraints. By employing the mean value theorem, the systems

Y.-J. Liu, S. Tong / Automatica 64 (2016) 7075

Fig. 3. (a) The trajectories of x1 and yd ; (b) phase portrait of z1 and z2 ; (c) the
trajectories of x1 and x2 .

are transformed as a strict feedback structure. Based on the transformed systems, a novel backstepping design is constructed with
the aid of the use of BLFs and we have proved that all the signals
in the closed-loop system are uniformly bounded and the tracking
error can converge to a small compact set without violating the
constraints. Finally, the performance of the approach can be illustrated through two simulation examples.
References
Chen, M., Ge, S. S., & Ren, B. (2011). Adaptive tracking control of uncertain MIMO
nonlinear systems with input constraints. Automatica, 47(3), 452465.
Chen, W., Hua, S., & Ge, S. S. (2014). Consensus-based distributed cooperative
learning control for a group of discrete-time nonlinear multi-agent systems
using neural networks. Automatica, 50(9), 22542268.
Ge, S. S., & Wang, C. (2002). Adaptive NN control of uncertain nonlinear purefeedback systems. Automatica, 38(4), 671682.
Ge, S. S., & Wang, C. (2004). Adaptive neural control of uncertain MIMO nonlinear
systems. IEEE Transactions on Neural Networks, 15(3), 674692.
Ge, S. S., Yang, C., & Lee, T. H. (2008). Adaptive predictive control using neural
network for a class of pure-feedback systems in discrete time. IEEE Transactions
on Neural Networks, 19(9), 15991614.
He, W., Ge, S. S., How, B. V. E., Choo, Y. S., & Hong, K. S. (2011). Robust adaptive
boundary control of a flexible marine riser with vessel dynamics. Automatica,
47(4), 722732.
He, W., Zhang, S., & Ge, S. (2014a). Adaptive control of a flexible crane system
with the boundary output constraint. IEEE Transactions on Industrial Electronics,
61(8), 41264133.
He, W., Zhang, S., & Ge, S. S. (2014b). Robust adaptive control of a thruster assisted
position mooring system. Automatica, 50(7), 18431851.
Kim, B. S., & Yoo, S. J. (2014). Approximation-based adaptive control of uncertain
non-linear pure-feedback systems with full state constraints. IET Control Theory
& Applications, 8(17), 20702081.
Krstic, M., Kanellakopoulos, I., & Kokotovic, P. V. (1995). Nonlinear and adaptive
control design. Wiley.
Krstic, M., & Kokotovic, P. V. (1996). Adaptive nonlinear output-feedback schemes
with MarinoTomei controller. IEEE Transactions on Automatic Control, 41(2),
274280.
Li, H., Gao, H., Shi, P., & Zhao, X. (2014). Fault-tolerant control of Markovian
jump stochastic systems via the augmented sliding mode observer approach.
Automatica, 50(7), 18251834.
Li, H., Pan, Y., & Zhou, Q. (2015). Filter design for interval type-2 fuzzy systems with
D stability constraints under a unified frame. IEEE Transactions on Fuzzy Systems,
23(3), 719725.
Li, Z. J., Xiao, H. Z., Yang, C. G., & Zhao, Y. (2015). Model predictive control
of nonholonomic chained systems using general projection neural networks
optimization. IEEE Transactions on Systems, Man, and Cybernetics: Systems,
45(10), 13131321.
Li, H., Yin, S., Pan, Y. N., & Lam, H. K. (2015). Model reduction for interval type-2
TakagiSugeno Fuzzy systems. Automatica, 61, 308314.
Liu, Y. J., Li, D. J., & Tong, S. (2014). Adaptive output feedback control for a class of
nonlinear systems with full-state constraints. International Journal of Control,
87(2), 281290.

75

Liu, Y. J., & Tong, S. (2015). Adaptive NN tracking control of uncertain nonlinear
discrete-time systems with nonaffine dead-zone input. IEEE Transactions on
Cybernetics, 45(3), 497505.
Mayne, D. Q., Rawlings, J. B., Rao, C. V., & Scokaert, P. O. (2000). Constrained model
predictive control: stability and optimality. Automatica, 36(6), 789814.
Nam, K., & Araposthathis, A. (1988). A model reference adaptive control scheme for
pure-feedback nonlinear systems. IEEE Transactions on Automatic Control, 33(9),
803811.
Ren, B., Ge, S. S., Tee, K. P., & Lee, T. H. (2010). Adaptive neural control for
output feedback nonlinear systems using a barrier Lyapunov function. IEEE
Transactions on Neural Networks, 21(8), 13391345.
Tang, Z. L. (2014). Adaptive neural network control of uncertain state-constrained
nonlinear systems. In IFAC world congress. Vol. 19, No. 1 (pp. 22792284).
Tee, K. P., & Ge, S. S. (2011). Control of nonlinear systems with partial state
constraints using a barrier Lyapunov function. International Journal of Control,
84(12), 20082023.
Tee, K. P., & Ge, S. S. (2012). Control of state-constrained nonlinear systems using
integral barrier Lyapunov functionals. In 51st IEEE conference on decision and
control. Maui, Hawaii, USA (pp. 32393244).
Tee, K. P., Ge, S. S., & Tay, E. H. (2009a). Barrier Lyapunov functions for the control
of output-constrained nonlinear systems. Automatica, 45(4), 918927.
Tee, K. P., Ge, S. S., & Tay, E. H. (2009b). Adaptive control of electrostatic
microactuators with bidirectional drive. IEEE Transactions on Control Systems
Technology, 17(2), 340352.
Tee, K. P., Ren, B., & Ge, S. S. (2011). Control of nonlinear systems with time-varying
output constraints. Automatica, 47(11), 25112516.
Tong, S. C., Sui, S., & Li, Y. (2015). Fuzzy adaptive output feedback control of MIMO
nonlinear systems with partial tracking errors constrained. IEEE Transactions on
Fuzzy Systems, 23(4), 729742.
Tong, S. C., Zhang, L. L., & Li, Y. M. (2015). Observed-based adaptive fuzzy decentralized tracking control for switched uncertain nonlinear large-scale systems
with dead zones. IEEE Transactions on Systems, Man, and Cybernetics: Systems,
http://dx.doi.org/10.1109/TSMC.2015.2426131. in press.
Wang, C. L., & Lin, Y. (2010). Adaptive dynamic surface control for linear
multivariable systems. Automatica, 46(10), 17031711.
Wang, C. L., & Lin, Y. (2012). Multivariable adaptive backstepping control: a norm
estimation approach. IEEE Transactions on Automatic Control, 57(4), 989995.
Yang, C., Ge, S. S., & Lee, T. H. (2009). Output feedback adaptive control of a class of
nonlinear discrete-time systems with unknown control directions. Automatica,
45(1), 270276.
Zhang, X., & Lin, Y. (2015). Adaptive output feedback control for a class of large-scale
nonlinear time-delay systems. Automatica, 52, 8794.
Zhang, Z., & Xu, S. (2015). Observer design for uncertain nonlinear systems with
unmodeled dynamics. Automatica, 51, 8084.
Zhou, B., Duan, G. R., & Lin, Z. (2011). A parametric periodic Lyapunov equation
with application in semi-global stabilization of discrete-time periodic systems
subject to actuator saturation. Automatica, 47(2), 316325.
Zhou, B., Li, Z. Y., & Lin, Z. (2013). Discrete-time l and l2 norm vanishment and low
gain feedback with their applications in constrained control. Automatica, 49(1),
111123.

Yan-Jun Liu received the B.S. degree in Applied Mathematics and the M.S. degree in Control Theory and Control Engineering from Shenyang University of Technology,
Shenyang, China, in 2001 and 2004, respectively. He received the Ph.D. degree in Control Theory and Control Engineering from Dalian University of Technology, Dalian,
China, in 2007.
He is currently a Professor with the College of Science,
Liaoning University of Technology. He is now an Associate
Editor of IEEE/CAA Journal of Automatica Sinica and Mathematical Problems in Engineering. His current research interests include adaptive fuzzy control, nonlinear control, neural network control,
reinforcement learning and optimal control.

Shaocheng Tong received the B.S. degree in Department of

Mathematics from Jinzhou Normal College, China, in 1982.
He received the M.S. degree in Department of Mathematics
from Dalian Marine University in 1988 and the Ph.D.
degree in control theory and control engineering from
Northeastern University, China, in 1997.
He is currently a Professor with the College of
Science, Liaoning University of Technology. He is now an
Associate Editor of IEEE Transactions on Systems, Man
and Cybernetics: Systems; International Journal of Fuzzy
Systems; Neurocomputing. His current research interests
include fuzzy and neural networks control theory and nonlinear control, adaptive
control, and system identification.