See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228625324

Hierarchical sliding mode control for a class of SIMO under-actuated systems

Article  in  Control and cybernetics · January 2008

DOI: 10.1504/IJMIC.2011.041779

CITATIONS READS

51 858

3 authors, including:

Jianqiang Yi Dongbin Zhao

Chinese Academy of Sciences Chinese Academy of Sciences
370 PUBLICATIONS   3,814 CITATIONS    328 PUBLICATIONS   4,205 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

StarCraft View project

All content following this page was uploaded by Dongbin Zhao on 21 May 2014.

The user has requested enhancement of the downloaded file.

 Control and Cybernetics
vol. 37 (2008) No. 1

Hierarchical sliding mode control for a class of SIMO

under-actuated systems∗
by
Dianwei Qian, Jianqiang Yi and Dongbin Zhao

Laboratory of Complex Systems and Intelligence Science,

Institute of Automation, Chinese Academy of Sciences
Beijing, 100190, P.R. China
e-mail: [email protected], [email protected], [email protected]

Abstract: A hierarchical sliding mode control approach is pro-

posed for a class of SIMO under-actuated systems. This class of
under-actuated systems is made up of several subsystems. Based
on this physical structure, the hierarchical structure of the sliding
surfaces is designed as follows. At first, the sliding surface of every
subsystem is defined. Then the sliding surface of one subsystem is
defined as the first layer sliding surface. The first layer sliding sur-
face is used to construct the second layer sliding surface with the
sliding surface of another subsystem. This process continues till the
sliding surfaces of the entire subsystems are included. According to
the hierarchical structure, the total control law is deduced by the
Lyapunov theorem. In theory, the asymptotic stability of the entire
system of sliding surfaces is proven and the parameter boundaries of
the subsystem sliding surfaces are given. Simulation results show the
feasibility of this control method through two typical SIMO under-
actuated systems.
Keywords: control approaches, control applications, sliding
mode control, under-actuated systems.

1. Introduction
Under-actuated systems are characterized by the fact that they have fewer ac-
tuators than the degrees of freedom to be controlled. They arise in extensive
applications such as robots, manipulators and industrial equipments. Their
dynamics often contains feedforward nonlinearities, non-minimum phase zero
dynamics, nonholonomic constraints, and couplings, which make their control
designs difficult (Spong, 1998). Recently, there has been an increasing attention
to the respective control problems in theory and in practice.
∗ Submitted: March 2007; Accepted: March 2008
160 D. QIAN, J. YI, D. ZHAO

In this study, we focus on a class of SIMO under-actuated systems. This class

is rather large, including series, parallel or rotary inverted pendulum(s) systems,
Pendubot, TORA, and so on. Such systems are often used for investigation
of various control methods and for education in various concepts. Numerous
control methods have been presented for them, such as energy-based control
(Fantoni et al., 2000, and Xin et al., 2004), passivity-based control (Alleyne,
1998), hybrid control (Zhang and Tarn, 2002, 2003), intelligent control (Yi et
al., 2002, and Lai et al., 1999), and partial feedback linearization techniques
(Spong, 1995). Other results concerning them can be found in Rubi et al.
(2002), Ortega et al. (2002), Fang et al. (2003), and Jung et al. (2004). Most
papers proposed a control law for a specified system. There is lack of general
results. In fact, a normal state space expression can depict this class. Thus, it
is possible to design a control law for this class rather than one control law for
one specified system.
Sliding mode control (SMC) is a powerful and robust nonlinear control
method (Bartoszewicz, 2000, and Gao et al., 1993). It provides a good can-
didate for the control design of this class. But designing a conventional single
layer sliding surface is not appropriate, because parameters of the sliding sur-
face cannot be calculated directly according to Hurwitz condition as in linear
systems (Wang et al., 2004). As far as physical structure is concerned, the class
considered can be divided into several subsystems. Based on this structure,
some control methods have been presented. Mon and Lin (Mon et al., 2002,
and Lin et al., 2005) proposed a hierarchical fuzzy sliding mode control scheme.
In the scheme, the controller parameters were modified by fuzzy logic. But they
did not consider the stability of the subsystem sliding surfaces. Lo (Lo et al.,
1998) designed a decoupled fuzzy sliding mode control law. By an intermedi-
ate variable, the whole system was decoupled into two levels. But this method
could not be applied to n-level control (n > 2). As in the method presented
by Mon and Lin, Lo also did not give a strict proof about the stability of the
subsystem sliding surfaces. Yi (Yi et al., 2005) developed a cascade sliding
mode control strategy. This controller was globally stable in the sense that all
signals involved were bounded. But some controller parameters often needed to
be switched to guarantee system stability. This might make it difficult to select
controller parameters. Wang (Wang et al., 2004) designed a hierarchical sliding
mode control law. The entire sliding surfaces were asymptotically stable. But
the control law could not be general for the under-actuated systems with more
subsystems than two. By considering the stability of subsystems, the param-
eters of the subsystem sliding surfaces in the above four papers were selected
as positive constants. But none of them gave the upper boundaries of these
parameters.
In this paper, a hierarchical sliding mode controller is developed for this class
of SIMO under-actuated systems. In this approach, such an under-actuated sys-
tem is divided into several subsystems according to its physical structure. The
sliding surface of every subsystem is defined. Then, the sliding surface of one
Hierarchical sliding mode control for SIMO under-actuated systems 161

subsystem is chosen as the first layer sliding surface. It is used to constructs

the second layer sliding surface with the sliding surface of another subsystem.
This process continues till all the subsystem sliding surfaces are included. Using
Lyapunov stability theorem, the total control law is derived. Theoretically, the
asymptotic stability of the entire system of sliding surfaces is proven. Further,
we prove that the parameters of the subsystem sliding surfaces should be con-
vergent to an open interval. To the best of the authors’ knowledge, the upper
boundaries of the parameters of the subsystem sliding surfaces are given for the
first time. The simulation results show the feasibility of this control strategy
through upswing control of a Pendubot system and stabilization control of a
series double inverted pendulum system.

2. Design of the hierarchical sliding mode control

Consider the state space expression of the class of SIMO under-actuated systems
with n subsystems in the following normal form:


 ẋ1 = x2
ẋ2 = f1 + b1 u




 3 = x4
ẋ



ẋ4 = f2 + b2 u (1)
 ..
.




ẋ = x2n

2n−1



ẋ2n = fn + bn u


Here, X = [x1 , x2 , . . . , x2n ]T is the state variable vector; fi and bi (i = 1, 2, . . . , n)

are the nonlinear functions of the state vector; and u is the single control input.
Equation (1) can express the class of systems with different n, fi and bi . If
n = 2, (1) can represent Pendubot, single inverted pendulum system and so on;
if n = 3, it can express a series or parallel double inverted pendulum system;
if n = 4, it can be considered as triple inverted pendulum system; and so on.
Based on physical structure, the class of under-actuated systems can be divided
into several subsystems. For example, a triple inverted pendulum system can
be divided into four subsystems: upper pendulum, middle pendulum, lower
pendulum, and cart. Such a system in (1) is made up of n subsystems. The ith
subsystem includes the state variables x2i−1 and x2i . Its state space expression
is represented by

ẋ2i−1 = x2i
. (2)
ẋ2i = fi + bi u

Define its sliding surface as

si = ci x2i−1 + x2i . (3)

162 D. QIAN, J. YI, D. ZHAO

Here, ci is a positive constant.

Differentiate si with respect to time t in (3), then, there exists
ṡi = ci ẋ2i−1 + ẋ2i = ci x2i + fi + bi u. (4)
Let ṡi = 0 in (4), the equivalent control of the ith subsystem is obtained as
ueqi = −(ci x2i + fi )/bi . (5)

According to (3) and (5), a variety of hierarchical sliding mode control laws
can be designed according to diverse combinations of the subsystem sliding
surfaces (Yi et al., 2005, and Wang et al., 2004). Other control methods can
also be combined with the hierarchical sliding mode method (Lin and Mon,
2005; Lo et al., 1998). In this paper, the hierarchical structure of the sliding
surfaces is designed according to the following description. The sliding surface
of one subsystem is chosen as the first layer sliding surface S1 . Then S1 is
used to construct the second layer sliding surface S2 with the sliding surface of
another subsystem. This process continues till all the subsystem sliding surfaces
are included. Without loss of generality, s1 is defined as S1 . The hierarchical
structure of the sliding surfaces is shown in Fig. 1.

Sn

Sn-1 sn

S2 s3
x2n-1 x2n
S1 s2
x5 x6

x1 x2 x3 x4

Figure 1. Hierarchical structure of the sliding surfaces

In Fig. 1, the ith layer sliding surface includes the entire information of the
other i − 1 layers sliding surfaces and the ith-subsystem sliding surface. As a
result, the ith layer sliding surface is defined as follows:
Si = λi−1 Si−1 + si . (6)
Here, λi−1 (i = 1, 2, . . . , n) is constant and λ0 = S0 = 0.
Correspondingly, the ith layer sliding mode control law should include the
information of the other i − 1 lower layers sliding mode control laws and the
Hierarchical sliding mode control for SIMO under-actuated systems 163

control law of the ith subsystem. Hence, the ith layer sliding mode control law
is defined as

ui = ui−1 + ueqi + uswi . (7)

Here, u0 = 0 and uswi (i = 1, 2, . . . , n) is the switching control of the ith layer

sliding surface.
The above control law (7) can be derived from the Lyapunov stability theo-
rem. The Lyapunov function of the ith layer sliding surface is selected as

Vi (t) = Si2 /2. (8)

Differentiating Vi with respect to time t in (8), we obtain from (6):

V̇i = Si Ṡi = Si (λi−1 Ṡi−1 + ṡi ). (9)

Further, (10) can be deduced from (6):

X i
i Y 
Si = aj sr . (10)
r=1 j=r

Here, aj = λj (j = 1, 2, . . . , i − 1) is constant; and ai = 1.

Hence, (11) is derived from (9) and (10):

hX i
i Y  i
V̇i = Si Ṡi = Si aj ṡr . (11)
r=1 j=r

From (4), (5), (7), and (11), (12) is obtained:

nX i
i h Y  io
V̇i = Si aj · (cr x2r + fr + br ui )
r=1 j=r
nX i
i h Y  X i
X io
= Si aj · b r · ueql + uswl
r=1 j=r l=1i l=1
l6=r
i hX
nX i Y
i  i i hX
X i
i Y  i o
= Si · aj br · ueql + aj br · uswl . (12)
l=1 r=1 j=r l=1 r=1 j=r
r6=l

By considering the stability of the ith layer sliding surface, let

Ṡi = −ki Si − ηi sgn Si , (13)

where ki and ηi are positive constants.

164 D. QIAN, J. YI, D. ZHAO

From (12) and (13), the switching control law of the ith layer can be obtained
as
Pi h P i  Qi  i
aj br ueql
i−1 l=1 r=1 j=r
X r6=l ki Si + ηi sgn Si
uswi = − uswl − i  i  − i  i . (14)
P Q P Q 
l=1 aj b r aj b r
r=1 j=r r=1 j=r

Let i = n in (7) and (14), then, there exists the hierarchical sliding mode control
law as follows:
n−1
X n
X
un = uswl + uswn + ueql
l=1 l=1
n
P n
Q 
aj br ueqr
r=1 j=r kn Sn + ηn sgn Sn
= Pn  Q
n  − Pn  Qn  . (15)
aj b r aj b r
r=1 j=r r=1 j=r

In (15), only the switching control of the last layer sliding mode controller
works. The switching controls of the other n-1 lower layers are merged. In
the reaching mode stage, when any state of a subsystem deviates from its own
subsystem sliding surface, the switching control of the last layer will drive it
back. Consequently, in the sliding mode stage, the system states keep sliding
on the last layer sliding surface. Moreover, the states of every subsystem keep
sliding on the subsystem sliding surface itself.

3. Stability analysis
Three theorems will be proven in this section. Theorem 1 is about the asymp-
totic stability of every layer sliding surface. Theorem 2 gives the stability anal-
ysis of the subsystem sliding surfaces. Further, we prove that the parameter of
the ith-subsystem sliding surface should be convergent to an open interval in
Theorem 3.
Theorem 1 Consider the class of under-actuated systems (1). If the control
law adopted is (15) and the ith layer sliding surface is defined as (6), then Si is
asymptotically stable.
Proof. Differentiate Vi (t) with respect to time t in (8), then, from (13), we
obtain
V̇i = Si Ṡi
= Si (−ki Si − ηi sgn Si )
= −ki Si2 − ηi |Si |. (16)
Hierarchical sliding mode control for SIMO under-actuated systems 165

Integrating both sides of (16) yields

Z t Z t
V̇i dτ = (−ki Si2 − ηi |Si |)dτ . (17)
0 0

Further,
Z t
Vi (0) = Vi (t) + (ki Si2 + ηi |Si |)dτ
0
Z t
> (ki Si2 + ηi |Si |)dτ .
0

Hence
Z t
lim (ki Si2 + ηi |Si |)dτ 6 Vi (0) < ∞. (18)
t→∞ 0

According to Barbalat’s lemma, there exists

lim (ki Si2 + ηi |Si |) = 0. (19)

t→∞

(19) means that lim Si = 0. Namely, the ith layer sliding surface Si is asymp-
t→∞
totically stable.

Theorem 2 Consider the class of under-actuated systems (1). If the control

law adopted is (15) and the ith-subsystem sliding surface si is defined as (3),
then si is asymptotically stable.

Proof. From Theorem 1, we have

lim Si = 0. (20)
t→∞

Let us prove that all the ith-subsystem sliding surfaces are asymptotically stable
by contradiction. It is assumed that si is not asymptotically stable, namely

lim si 6= 0. (21)
t→∞

Calculating the limit of both sides of (10) yields

X i
i Y 
lim Si = lim aj sr
t→∞ t→∞
r=1 j=r
i h
X i
Y  i
= lim aj sr
t→∞
r=1 j=r
i h
X i
Y  i
= aj lim sr =6 0. (22)
t→∞
r=1 j=r
166 D. QIAN, J. YI, D. ZHAO

(22) contradicts (20) that we have obtained from Theorem 1, so that the
assumption (21) is false and the case opposite to (21) must be true. Therefore,
the sliding surfaces of the entire set of subsystems are asymptotically stable.

Theorem 3 For the class of under-actuated systems (1), adopting the control
law (15), defining the ith-subsystem sliding surface (3) and assuming that all
the state variables are equivalent infinitesimals of each other at a certain neigh-
borhood of the origin, we have that the boundary of the parameter ci should be
0 < ci < | lim (fi /x2i )|, where X is the state vector.
X→0

Proof. When the states of the ith subsystem keep sliding on the subsystem
sliding surface itself, the following equation can be obtained:


 si = ci x2i−1 + x2i = 0 (a)
ṡi = ci ẋ2i−1 + ẋ2i = 0 (b)

(23)
 ẋ2i−1 = x2i
 (c)
ẋ2i = fi + bi ueqi . (d)


1) Lower boundary of ci
From (23a) and (23c), there exists

si = ci x2i−1 + ẋ2i−1 = 0. (24)

The eigenvalue of (24) should be negative for guaranteeing the stability of the
ith-subsystem sliding surface. Thus, the lower boundary of ci is ci > 0.
2) Upper boundary of ci
From (23b) and (23d), there exists

ṡi = ci ẋ2i−1 + ẋ2i = ci x2i + fi + bi ueqi = 0. (25)

Further, we have

ci = |(fi + bi ueqi )/x2i | 6 (|fi | + |bi ueqi |)/|x2i |. (26)

When the states of the ith subsystem keep sliding on its own sliding surface
and converge to a certain neighborhood of the origin, this subsystem can be
treated as an autonomous system. Hence, we have the following inequality:

ci 6 |fi /x2i |. (27)

According to the limit concept of a multivariable function, ci0 = | lim (fi /x2i )|
X→0
can be calculated through the assumption that all the state variables are equiv-
alent infinitesimals of each other at a certain neighborhood of the origin.
From part 1 and part 2, the boundary of the parameter ci should be 0 <
ci < ci0 .
Hierarchical sliding mode control for SIMO under-actuated systems 167

Theorem 3 is a necessary condition about the parameter ci . If a self-adaptive

control law is adopted, the value of ci may not be in the open interval (0, ci0 )
when the system states are far from the origin. But the parameter ci will
converge to this open interval as the system states converge to the origin. Al-
though this theorem does not permit to calculate a precise value, it points out
a direction for selecting the parameter ci . Further, this theorem does not mean
any number between 0 and ci0 could be selected as the value of the parameter
ci . The reason is that the different subsystem sliding surface parameters are
coupled with each other in the controller. If one parameter is selected, other
parameters, which still satisfy Theorem 3, have to be selected after trial and
error by simulations. How to determine all the controller parameters directly
may constitute another important research direction.

4. Simulation results
Pendubot and series double inverted pendulum system are two typical under-
actuated systems, which are often used to verify the feasibility of new control
methods. Both of their dynamics have the similar expressions as (1) with differ-
ent fi , bi , and n. In this section, the presented control method will be applied to
upswing control of a Pendubot system and stabilization control of a series dou-
ble inverted pendulum system. The demonstrations will show that this control
method is feasible.

4.1. Pendubot
The Pendubot system shown in Fig. 2 is made up of two subsystems: link 1
with an actuator (subscript 1) and link 2 without an actuator (subscript 2).
The control objective of its upswing control is to swing up the Pendubot from
the stable pending equilibrium point to its uppermost unstable equilibrium point
and to stabilize at the uppermost unstable equilibrium point (Zhang and Tarn,
2002 and 2003; Wang et al., 2004).
The symbols in Fig. 2 are defined as follows: θ1 is the angle of the link 1
with respect to the horizontal line; θ2 is the angle of the link 2 with respect to
the link 1; mi , li and lci are the mass, length and distance to the centre of mass
of link i, where i = 1, 2; τ1 is the control torque. Let n = 2 in (1), then, the
state space expression of the Pendubot system can be written as


 ẋ1 = x2
ẋ2 = f1 + b1 u

(28)

 ẋ3 = x4
ẋ4 = f2 + b2 u.


Here, x1 = θ1 − π/2) is the angle of the link 1 with respect to the vertical line;
x3 = θ2 is the angle of the link 2 with respect to the link 1; x2 is the angular
velocity of the link 1; x4 is the angular velocity of the link 2; u= τ1 is the single
168 D. QIAN, J. YI, D. ZHAO

m2
l2

l1 T2
m1
lc1 lc2

T1
W1 x

Figure 2. Structure of the Pendubot system

control input; the expressions for f1 , f2 , b1 and b2 are shown in Wang et al.
(2004).
For comparisons, the parameters of the Pendubot are selected as q1 =0.0308kg
· m2 , q2 =0.0106kg · m2 , q3 =0.0095kg · m2 , q4 =0.2086kg · m, q5 =0.0630kg · m,
and the gravitational acceleration g=9.81m · s−2 which have appeared in Zhang
and Tarn (2002 and 2003), Wang et al. (2004). According to Theorem 3, the
boundaries of c1 and c2 are calculated as 66.97 and 68.68, whose expressions are
given as
(
c10 = g|(q3 q5 − q2 q4 )/(q1 q2 − q32 )|
c20 = g|[q5 (q1 + q3 ) − q4 (q2 + q3 )]/(q1 q2 − q32 |.

At last, the parameters of the hierarchical sliding mode controller are chosen as
c1 =8.00, c2 =2.80, a1 =2.50, k2 =1.20, and η2 =0.10, after trial and error. The ini-
tial state vector X0 is [−π, 0, 0, 0]T and the desired state vector Xd is [0, 0, 0, 0]T .
Fig. 3 shows the angular curves. As Fig. 3 shows, this control strategy can
realize the control objective in about 4 seconds. Fig. 4 displays the control
torque. We can see from it that the largest control torque is about 4Nm.
Fig. 5 shows the entire sliding surfaces. As we have proven, not only every
layer sliding surface is asymptotically stable, but also the sliding surfaces of the
two subsystems possess the asymptotic stability.
By considering the property of equilibrium points, the control law in Wang
et al. (2004) needed to be adjusted for different under-actuated systems. In
practice, this may lead to a confusion. Further, Wang’s method could only
be applied for the under-actuated systems with two subsystems. Hence, the
method here presented is more general. Compared with Wang’s results, the
angular curves and the control torque are smoother. During the future physical
experiments, this advantage could decrease mechanical abrasions. The maxi-
mum of the control torque in Wang’s results exceeded 4Nm, which was a little
Hierarchical sliding mode control for SIMO under-actuated systems 169

T1, rad
0

-1

-2
3

2
T 2, rad

-1
0 2 4 6 8
t, s
Figure 3. Angles of the Pendubot

4
3
2
u, Nm

1
0
-1
0 2 4 6 8
t, s
Figure 4. Control torque

15

0
s1 and s2

s1=S1
-15
s2

-30
15
0
-15
-30
S2

-45
-60
-75
0 2 4 6 8
t, s
Figure 5. Sliding surfaces
170 D. QIAN, J. YI, D. ZHAO

more than the performance shown in Fig. 4. But the presented control method
was not as robust as Wang’s for upswing control of Pendubot system. Wang’s
controller resisted a periodical disturbance of y = 2 sin 10t and a step distur-
bance of 1 radian, which were out of reach for the present controller. In our
simulations, we found it could resist a periodical disturbance of y = 0.4 sin 6πt
and a step disturbance of 0.2 radian.

4.2. Series double inverted pendulum system

The series double inverted pendulum system is made up of double inverted
pendulums on a moving cart as is shown in Fig. 6. This system contains three
subsystems: the upper pendulum, the lower pendulum, and the cart. The
control objective of its stabilization control is to balance both of the pendulums
upright and to put the cart to the rail origin by moving the cart (Lin and Mon,
2005).

ș2

ș1

Figure 6. Structure of the series double inverted pendulum system

The symbols in Fig. 6 are defined as follows: θ1 is the lower pendulum angle
with respect to the vertical line; θ2 is the upper pendulum angle with respect
to the vertical line; x is the cart position with respect to the origin; u is the
control force. Let n = 3 in (1), the state space expression of the series double
inverted pendulum system is described below:


 ẋ1 = x2
ẋ2 = f1 + b1 u




ẋ3 = x4

(29)

 ẋ4 = f2 + b2 u
ẋ5 = x6




ẋ6 = f3 + b3 u.


Here, x1 = θ1 ; x3 = θ2 ; x5 = x; x2 is the angular velocity of the lower pendulum;

Hierarchical sliding mode control for SIMO under-actuated systems 171

x4 is the angular velocity of the upper pendulum; x6 is the velocity of the cart;
u is the single control input; fi and bi (i = 1, 2, 3) are given in Lo et al. (1998).
For simulative comparisons, the structural parameters of the series double
inverted pendulum system are chosen as the cart mass M =1kg, the lower pen-
dulum mass m1 =1kg, the upper pendulum mass m2 =1kg, the lower pendulum
length l1 =0.1m, the upper pendulum length l2 =0.1m, the gravitational acceler-
ation g=9.81m · s−2 , which have appeared in Lin and Mon (2005). According
to Theorem 3, the boundaries of c1 , c2 and c3 are calculated as 294.39, 98.31
and 11.44, whose expressions are given as
A2 (B/3 − m2 l2 /4)
  
 
c10 = g  (m2 /4 − A/3)(B 2 − AC) − m2 (B − Al1 )2 /4 

  




A2 (C − Bl1 )/2

  
 
c20 = g 
2 2


 l 2 [(m 2 /4 − A/3)(B − AC) − m 2 (B − Al 1 ) /4] 
  

  AB(B/3 − m2 l1 /4) + A(Cm2 − Bm2 l1 )/2 
c30 = g  (m2 /4 − A/3)(B 2 − AC) − m2 (B − Al1 )2 /4 

  


Here, A = M + m1 + m2 , B = m1 l1 /2 + m2 l1 and C = m1 l1 l1 /3 + m2 l2 l2 . The

controller parameters are chosen as c1 =184.26, c2 =15.96, c3 =0.72, a1 = -0.06,
a2 =0.45, k3 =1.50, and η3 =0.02, after trial and error. The initial state vector X0
and the desired state vector Xd are [π/6, 0, π/18, 0, 0, 0]T and [0, 0, 0, 0, 0, 0]T ,
which are the same as Lin and Mon (2005), Lo et al. (1998).
Fig. 7 shows the angular curves and the positional curve. The maximum
distance of the cart with respect to the origin is about 0.6 m. Fig. 8 displays
the control force applied to the cart.
Fig. 9 shows the curves of the subsystem sliding surfaces. We can see from it
that the subsystem sliding surfaces are asymptotically stable as has been proven
in Theorem 2. Fig. 10 displays the curves of the three layer sliding surfaces. It
shows that every layer sliding surface can converge to zero as has been proven
in Theorem 1.
Lo’s control method could only realize two-level control. Consequently, that
controller could only balance the double pendulums, but it failed to put the
cart to the origin at the same time. Compared with Lo’s results, our control
objective is more difficult. In Lin’s method, the controller parameters were
modified by fuzzy logic. But the fuzzy rules were difficult to define. Moreover,
the use of fuzzy logic made it difficult to analyze system stability. Compared
with Lin’s performance, the curves are smoother and the cart moves along a
shorter distance in Fig. 7, although a bigger control force in Fig. 8 is needed in
the beginning of the simulation, exceeding 200N. For most physical systems, it
is very difficult to offer such a force or it is too expensive to offer it. On the
other hand, such a force may make the input saturated, which may influence
the reliable operation and acceptable performance of the control system. This
weak point will restrict the control method in practice. Thus, solving of this
issue is the problem to be addressed in future investigations.
172 D. QIAN, J. YI, D. ZHAO

0.6
0.0

T1, rad
-0.6
-1.2
0.5

T2, rad
0.0
-0.5
-1.0
0.4
0.0
x, m

-0.4
-0.8
0 3 6 9
t, s

Figure 7. Pendulum angles and cart position

200

100
u, N

-100
0 3 6 9
t, s

Figure 8. Control force

100
0
s1

-100
-200
10
0
s2

-10
-20
2
0
s3

-2
-4
0 3 6 9
t, s

Figure 9. Sliding surfaces of the three subsystems

Hierarchical sliding mode control for SIMO under-actuated systems 173

100
0

S1
-100
-200
4
0
S2
-4
-8
1
0
S3

-1
-2
0 3 6 9
t, s

Figure 10. Three layer sliding surfaces

Fig. 11 shows the stabilization domain of the initial angles of the two pendu-
lums, for which the proposed hierarchical sliding mode controller can stabilize
the pendulum system. Here, the horizontal axis and the vertical axis stand
separately for the initial angles of the lower pendulum and the upper pendulum
in degree. The initial angle of the lower pendulum is selected every 5˚ from
-40˚ to 40˚. For such a selected angle of the lower pendulum, the maximum
angle of the upper pendulum that can be stabilized is plotted by the black filled
triangle (N) in Fig. 11. The initial values of the other state variables x2 , x4 ,
x5 , x6 are all fixed to zeros. During the simulations, the complete stabilization
time is within 6.0s and the cart position is in the interval [-1m, 1m].

50
40
30
Upper pendulum angle

20
180x3/S (deg)

10
0
-10
-20
-30
-40
-50
-50 -40 -30 -20 -10 0 10 20 30 40 50
Lower pendulum angle
180x1/S (deg)

Figure 11. Stabilization domain

174 D. QIAN, J. YI, D. ZHAO

5. Conclusions
The hierarchical sliding mode control scheme has been developed for a class
of SIMO under-actuated systems. According to the physical structure of the
class, the hierarchical sliding surfaces are designed. Using Lyapunov theorem,
the total control law is derived. The asymptotic stability of the entire system of
sliding surfaces is proven theoretically. To the best of the authors’ knowledge,
the upper boundaries on the parameter of the subsystem sliding surfaces are
given for the first time. In the simulation examples, performance of this control
method is demonstrated through upswing control of a Pendubot system and
stabilization control of a series double inverted pendulum system. The two
systems belong to the class considered, with different numbers of subsystems.
The simulation results show the controller validity and generalization. This
method can also be used to other under-actuated systems in the class besides
the two demonstrated systems, like an overhead crane system, an Acrobot and
so on. But there still exist some weak points in the presented control method,
such as how to select the controller parameters, how to deal with the problem
of the input saturation, and so on. The extensions and the physical experiment
(Liu et al., 2005) are still the future issues.

Acknowledgement
This work was partly supported by the NSFC Projects under Grant No. 60575047
and 60621001, the National 863 Program (No.2007AA04Z239), the Outstanding
Overseas Chinese Scholars Fund of Chinese Academy of Sciences (No. 2005-
1-11), and the Joint Laboratory of Intelligent Sciences & Technology (No.
JL0605), China.

References
Alleyne, A. (1998) Physical insights on passivity-based TORA control de-
signs. IEEE Transactions on Control Systems Technology 6, 436–439.
Bartoszewicz, A. (2000) Chattering attenuation in sliding mode control sys-
tems. Control and Cybernetics 29, 585-594.
Fang, Y., Dixon, W.E., Dawson, D.M. and Zergeroglu, E. (2003) Non-
linear coupling control laws for an underactuated overhead crane system.
IEEE/ASME Transactions on Mechatronics 8, 418–423.
Fantoni, I., Lozano, R. and Spong, M.W. (2000) Energy based control of
the pendubot. IEEE Transactions on Automatic Control 45, 725–729.
Gao, W. and Hung, J.C. (1993) Variable structure control of nonlinear sys-
tems: a new approach. IEEE Transactions on Industrial Electronics 40,
45–55.
Jung, S. and Wen, J.T. (2004) Nonlinear model predictive control for the
swing-up of a rotary inverted pendulum. ASME Journal of Dynamic Sys-
tems, Measurement and Control 120, 666–673.
Hierarchical sliding mode control for SIMO under-actuated systems 175

Lai, X., She, J.H., Ohyama, Y. and Cai, Z. (1999) Fuzzy control strategy
for acrobots combining model-free and model-based control. IEE Proceed-
ings - Control Theory and Applications 146, 505–510.
Lin, C.M. and Mon, Y.J. (2005) Decoupling control by hierarchical fuzzy
sSliding mode controller. IEEE Transactions on Control Systems Tech-
nology 13, 593–598.
Liu, D., Yi, J., Zhao, D. and Wang, W. (2005) Adaptive sliding mode fuzzy
control for a two-dimensional overhead crane. Mechatronics 15 505–522.
Lo, J.C. and Kuo, Y.H. (1998) Decoupled fuzzy sliding-mode control. IEEE
Transactions on Fuzzy Systems 6, 426–435.
Mon, Y.J. and Lin C.M. (2002) Hierarchical fuzzy sliding-mode control. Pro-
ceedings of 2002 IEEE International Conference on Fuzzy Systems 1, 656–
661.
Ortega, R., Spong, M.W., Gomezestern, F. and Blanhenstein, G.
(2002) Stabilization of a class of underactuated mechanical systems via
interconnection and damping assignment. IEEE Transactions on Auto-
matic Control 47, 1218–1233.
Rubi, J., Rubio, A. and Avello, A. (2002) Swing-up control problem for a
self-erecting double inverted pendulum. IEE Proceedings - Control Theory
and Applications 149, 169–175.
Spong, M.W. (1998) Underactuated Mechanical Systems. Lecture Notes in
Control and Information Sciences 230, 135–150.
Spong, M.W. (1995) The swing up control problem for the acrobot. IEEE
Control Systems Magazine 15, 49–55.
Wang, W., Yi, J., Zhao, D. and Liu, D. (2004) Design of a stable sliding-
mode controller for a class of second-order underactuated systems. IEE
Proceedings - Control Theory and Applications 151, 683–690.
Xin, X. and Kaneda, M. (2004) New Analytical Results of the Energy Based
Swinging up Control of the Acrobot. Proceedings of the 43rd IEEE Con-
ference on Decision and Control, 1, 704–709.
Yi, J., Yubazaki, N. and Hirota, K. (2002) A new fuzzy controller for sta-
bilization of parallelt-type double inverted pendulum system. Fuzzy Sets
and Systems 126, 105–119.
Yi, J., Wang, W., Zhao, D. and Liu, X. (2005) Cascade sliding-mode con-
troller for large-scale underactuated systems. Proceedings of 2005 IEEE -
RSJ International Conference on Intelligent Robots and Systems, 301–306.
Zhang, M. and Tarn, T.J. (2002) Hybrid control of the pendubot. IEEE -
ASME Transactions on Mechatronics 7, 79–86.
Zhang, M. and Tarn, T.J. (2003) Hybrid switching control strategy for non-
linear and underactuated mechanical systems. IEEE Transactions on Au-
tomatic Control 48, 1777–1782.
View publication stats