Inverted Pendulum

Available online at www.sciencedirect.
com
Journal of the Franklin Institute 351 (2014) 22612282

www.elsevier.com/locate/jfranklin
Design and implementation of a new sliding mode

controller on an underactuated wheeled
inverted pendulum
Zhao-Qin Guoa,b, Jian-Xin Xua,b,n, Tong Heng Leea,b
a
Graduate School for Integrative Sciences and Engineering, National University of Singapore,
Singapore 117456, Singapore
b
Department of Electrical and Computer Engineering, National University of Singapore,
4 Engineering Drive 3, Singapore 117576, Singapore
Received 30 June 2012; received in revised form 21 December 2012; accepted 7 February 2013
Available online 15 February 2013
Abstract
In this paper, a sliding mode controller (SMC) is proposed for control of a wheeled inverted
pendulum (WIP) system, which consists of a pendulum and two wheels in parallel. The control
objective is to use only one actuator to perform setpoint control of the wheels while balance the
pendulum around the upright position, which is an unstable equilibrium. When designing the SMC
for the WIP system, various uncertainties are taken into consideration, including matched
uncertainties such as the joint friction, and unmatched uncertainties such as the ground friction,
payload variation, or road slope. The SMC proposed is capable of handling system uncertainties and
applicable to general underactuated systems with or without input coupling. For switching surface
design, the selection of the switching surface coefcients is in general a sophisticated design issue
because those coefcients are nonafne in the sliding manifold. In this work, the switching surface
design is transformed into a linear controller design, which is simple and systematic. By virtue of the
systematic design, various linear control techniques, such as linear quadratic regulator (LQR) or
linear matrix inequality (LMI), can be incorporated in the switching surface design to achieve
optimality or robustness for the sliding manifold. To further improve the WIP responses, the design
of reference signals is addressed. The reference position for the pendulum is adjusted according to the
actual equilibrium of the pendulum, which depends on the size of the friction and slope angle of the
traveling surface. A smooth reference trajectory for the setpoint of the wheel is applied to avoid
n
Corresponding author at: Graduate School for Integrative Sciences and Engineering, National University of
Singapore, Singapore 117456, Singapore.
E-mail addresses: guozhaoqin@gmail.com (Z.-Q. Guo), elexujx@nus.edu.sg (J.-X. Xu),
eleleeth@nus.edu.sg (T.H. Lee).
0016-0032/$32.00 & 2013 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jfranklin.2013.02.002
2262
Z.-Q. Guo et al. / Journal of the Franklin Institute 351 (2014) 22612282
abrupt jumps in the system responses, meanwhile the reaching time of the switching surface can be
reduced. The effectiveness of the SMC is validated using intensive simulations and experiment
testings.
& 2013 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
1. Introduction
Systems that have fewer control inputs than the degrees of freedom (DOF) to be
controlled are dened as underactuated systems. Control of underactuated systems is a
popular research topic due to its wide range of applications in robotics, underwater
vehicles, aerospace vehicles, etc. [1,2]. From practical concerns such as cost reduction or
weight reduction, many systems are designed to be underactuated. Some systems become
underactuated when actuator failure occurs. As benchmark examples of nonlinear and
underactuated systems, the cart-pendulum is often used to demonstrate and verify the
effectiveness of control algorithms.
In recent years, the control of WIP systems has attracted attentions from both researchers
and engineers. The well known commercial product, two-wheeled SEGWAY, is a popular
personal transporter. For research and education purposes, prototypes of autonomous WIPs
have been designed in universities and research institutes [39]. The WIP usually consists of
two actuated wheels in parallel and an unactuated inverse pendulum. The control objective of
the WIP is to perform position or velocity control of the wheels while stabilize the pendulum
around the upright position that is an unstable equilibrium point. The WIP developed in [39]
belongs to underactuated systems without input coupling [1]. The control input acts on the
wheels only, while the balancing of the pendulum is achieved by the wheel motion. The
mathematic model of the WIP system in [39] is essentially the same as that of the classic cartpendulum system, therefore the control methods designed for cart-pendulum system can be
directly applied to WIP system without input coupling.
Due to the difference in mechanical conguration, underactuated WIPs can be classied
into the class without input coupling where the actuator is mounted on the wheel (class A),
and the class with input coupling where the actuator is mounted on the pendulum or
chassis (class B). Class A is more complex in mechanical construction but easier in
controller design owing to the absence of input coupling. In contrast, class B is easier in
mechanical construction but more challenging in controller design due to the input
coupling between the wheel and the pendulum. Since the existing works mostly focus on
studying control of underactuated systems without input coupling, this work is devoted to
the development and control of a WIP with input coupling. When building the prototype
of the WIP, the motor shaft coupler is xed at the center of the wheels and the motor
housing is rigidly connected to the inverse pendulum, thus the torque generated by the
motor directly acts on both the wheels and the pendulum with the same size but opposite
directions, which results in the input coupling of the WIP system.
Considering that various uncertainties exist in the WIP system, for instance, the joint
and the ground frictions, the slope angle of the traveling surface, robustness is addressed in
the control system design. SMC is a well known robust control approach for systems with
model uncertainties and external disturbances and has been studied for control of wheeled
inverted pendulum and similar underactuated mechanical systems [1018]. SMC utilizes a
discontinuous control law to drive system state trajectory into a designer specied
2263
switching surface and to maintain the system state trajectory on this surface for all the
subsequent time. In standard SMC design for full actuated system, it is straightforward to
conclude the convergence of the system states from the switching surface equation when
system is in the sliding mode. However, for underactuated system, the standard SMC
design and stability analysis are not applicable because the system has fewer inputs than
the independent variables to be controlled. As a result, a nonlinear sliding manifold or in
general an internal dynamics must be stabilized by the proper selection of the switching
surface coefcients.
In [1013], coupled sliding mode control laws along with a linear coupled sliding surface
are proposed for controlling of the underactuated system, where the coupled linear
switching surface is designed by incorporating multiple independent state variables into a
scalar switching surface. The single actuator can thus be used to manipulate the scalar
switching surface. Conclusions regarding the convergence of the states, however, cannot be
drawn directly from the convergence of the switching surface due to the presence of the
sliding manifold or the internal dynamics, thus the stability of the sliding motion governed
by the sliding manifold should be further investigated, namely, the switching surface design
is required. In [1012], stable sliding manifolds are obtained with appropriately selected
switching surface coefcients. In [13], a stabilizing control law is assumed existing. In [14],
an SMC design based on the cascade normal form [1] is proposed, and the validity holds
under certain assumptions. However, the WIP studied in our work does not meet these
assumptions. Second-order SMC designs for underactuated systems are discussed in
[1518]. The design of second-order SMC requires that the derivative of the dened sliding
variable is known. In [1517], the SMC design requires that the derivatives of all system
states are known. However, in this work, the derivatives of the velocity states are not
available because the WIP system is in the presence of both parametric and external
uncertainties. In [19], a new nonlinear sliding surface is proposed to control a class of
nonminimum phase underactuated mechanical systems. The structure of the nonlinear
sliding surface is more complex than that of a linear sliding surface.
Design of optimal SMC algorithms has attracted particular interests recently [2128]. In
a typical SMC design, stability is the only concern in the switching surface design. The
optimal SMC design aims at achieving both robust and optimal control, thus shows the
superiority in practical applications. However, no results about optimal SMC design for
underactuated system have been shown in existing works. Thus, it would be meaningful
and interesting to explore optimal or suboptimal SMC designs for underactuated systems.
In this paper, for the easiness in real-time implementation, an SMC along with a linear
switching surface is proposed for setpoint control of the WIP system. The linear switching
surface is constructed by combining the two states of the wheel and two states of the
pendulum in a linear form [1013], which brings four coefcients associated with the four
states. The SMC law is derived using Lyapunov theory, which guarantees the nite
reaching time of the switching surface and leads to a sliding manifold with all the matched
uncertainties rejected. In the sliding mode, the sliding motion is determined by the four
coefcients, however, in a complex and highly nonlinear form. Therefore, it is difcult to
directly choose or tune the coefcients to achieve the desired sliding motion. To simplify
the switching surface design, the sliding manifold is linearized around the desired
equilibrium point of the pendulum. Through a mathematical transformation, it is shown
that the linearized sliding manifold is equivalent to a normal linear system that is under a
full state linear feedback control with the freedom in choosing feedback gains. Now the
2264
switching surface design becomes a nominal linear controller design, which is simple,
systematic, and furthermore provides one extra degree of freedom in control. In this work,
this degree of freedom is utilized to implement optimal or robust linear control techniques.
Two alternative methods are adopted for the nominal linear controller design. One is based
on LQR method, which leads to a stable sliding manifold that also exhibits optimality in
terms of fast tracking convergence and control cost. The other is based on LMI method
and the resulting sliding manifold exhibits robustness with respecting to various
unmatched uncertainties. The resulting sliding manifold exhibits desirable properties
besides stability, such as optimality and robustness. The existing works on SMC design for
underactuated systems hitherto only focus on the stabilization of the sliding manifold.
The main contributions of this paper are summarized as follows.
(i) An SMC is proposed to control an underactuated WIP system in the presence of both
matched and unmatched uncertainties. The proposed control methods and the
obtained results can be extended to general underactuated systems with or without
input coupling.
(ii) To make the control algorithm be simple and implementable, the SMC uses a linear
switching surface and the switching surface design is transformed into a nominal linear
control design, which is simple, systematic and furthermore provides one extra degree
of freedom in control. By utilizing the extra degree of freedom, various linear control
techniques can be incorporated in the SMC design. Thanks to the easiness in tuning of
the controller parameters, the SMC is successfully implemented on the WIP system.
Satisfactory responses are obtained for both regulation and setpoint control tasks.
(iii) The particular characteristics of the underactuated WIP system with input coupling
are investigated, according to which, references for both the wheel and the pendulum
are designed to improve system performance.
Throughout this paper, a function F l1 ,l2 , . . . ,ln will be written as F, where

l1 ,l2 , . . . ,ln can be parameters or variables.
The remainder of this paper is organized as follows. In Section 2, the underactuated
system model is given. In Section 3, SMC with linear switching surface design is detailed.
In Section 4, switching surface design is discussed. In Section 5, simulation based case
studies are presented. Section 6 presents the implementation of the proposed SMC on the
WIP platform. Conclusions are drawn in Section 7.
2. Problem formulation
Fig. 1 shows the prototype of the WIP. Since the lateral stability of the WIP is
guaranteed, in this work we focus on longitudinal control of the WIP.
2.1. WIP model
Fig. 2 shows the model of the WIP. The wheel motion is dened along the road surface.
_ respectively, with rightward
The wheel displacement and velocity are denoted by x and x,
as positive direction. y is the tilting angle of the pendulum with the upright position as zero
and clockwise rotation as positive direction. y_ is angular velocity of the pendulum. j is the
2265
Fig. 1. Prototype of the WIP.
Fig. 2. Model of the WIP system.
slope angle of the inclined road, for traveling on at surface, j 0. fr is the ground
friction. t is the torque generated by the motor and acting on the wheel with clockwise
rotation as positive direction, which is also the control input u to the system. Note that the
motor driving the wheel is directly mounted on the pendulum, there is a reaction torque t
applied to the pendulum. tf is the joint friction, which also acts on both the wheel and the
pendulum as tf and tf , respectively. Other system parameters are as
mw 1:551 kg: the mass of the wheels;
mp 1:6 kg: the mass of the pendulum;
Iw 0:005 kg m2 : the rotation inertia of the wheels;
Ip 0:027 kg m2 : the rotation inertia of the pendulum;
r 0.08 m: the radius of the wheel;
l 0.13 m: the distance between Center of Gravity (COG) of the pendulum and the
center of the wheel;
g 9.81 m/s2: the acceleration of gravity.
2266
The following assumptions are made in the formulation of the WIP dynamic model. (i)
There is no rolling slip between the wheels and the ground. (ii) The wheels are always in
contact with the ground.
Lagrangian mechanics method is used to derive the mathematical model of the WIP
system [20], which leads to a second-order nonlinear model given by
2
_
ax bym
p l siny jy sin jmp mw g
1
t tf rf r ,
r
bx cym
p lg sin y ttf ,
1
2
where a mw mp Iw =r2 , b mp l cosx3 j and c Ip mp l 2 .

Remark 1. From the system viewpoint, for WIP with input coupling, the control input t
exists in both the wheel and the pendulum motion Eqs. (1) and (2). While for WIP without
input coupling, t only exists in motion equation of the wheel subsystem (1).
2.2. Control objective
The control objective is to achieve setpoint control of the wheel, while balance the
_ T and
_ y, y
pendulum at an equilibrium (y ye , y_ 0). Dene x x1 , x2 , x3 , x4 T x, x,
T
the reference signal for x is chosen as r xr , vr ,yr , 0 with x_ r vr . We obtain the error
states as e e1 , e2 , e3 , e4 T xr x1 xr , x2 vr , x3 yr , x4 T . Now the control objective is to ensure the convergence of e. The error dynamic model of the WIP is obtained as
e_ ge geu dm e,t du e,t,
where g is the system nonlinear term, dm is the lumped matched uncertainties, du is the
lumped unmatched uncertainties. We have
ge e2 , Z1 e, e4 , Z2 eT ,
ge 0, g1 e, 0, g2 eT ,
dm e,t tf ,
du e,t 0, du1 e,t, 0, du2 e,tT ,
where
mp l
cmp mw g sin j
ce24 sine3 yr jbg sine3 yr
_v r ,
2
acb
acb2
mp l
bmp mw g sin j
Z2
be24 sine3 yr j ag sine3 yr
,
2
acb
acb2
1 c
b
g1
,
r acb2 acb2
1 b
a
g2
,
r acb2 acb2
c
fr ,
du1
acb2
b
du2
fr ,
acb2
and b mp l cose3 yr j.
Z1
2267
Remark 2. In the following work, control design and stability analysis are based on the
expression of dynamics model in Eq. (3), which can also be used to describe general
underactuated systems. Thus, the SMC designed for the WIP system is extendable to
general underactuated systems with or without input coupling.
2.3. Trajectory planning
The WIP is supposed to reach a desired position xd and stop there. However, using a
step function as the desired trajectory for the WIP would cause the desired traveling
velocity to be discontinuous at the beginning of the motion. The discontinuousness would
cause considerable jerk and pendulum swinging [6] in the response. High-order
polynomials are used for computing smooth trajectory [6]. In this work, we simply use a
linear segment and two parabolic blends to construct a smooth trajectory for the WIP,
which also yields a smooth reference signal for the wheel velocity. The reference inputs are
computed by the following equations and shown in Fig. 3:
8v
m
>
t
0otot1 ,
>
>
> t1
>
>
< vm
t1 rtrt2 ,
4
vr t
vm
>
vm
tt2 t2 rtrt3 ,
>
>
>
t3 t2
>
>
:0
t3 rtrts ,
Z
xr t
vr ds,
and xr ts xd , where xd is the setpoint.
Ref for Wheel Position (m) / Vel (m/s)
Remark 3. With the planned trajectory, the absolute value of initial e1 and e2 are zero, and
this is desirable in the SMC design. First, for system under feedback control, like SMC,
large initial errors could result in a large control input at the beginning, which may lead to
unstable responses. Second, later it can be found that the WIP system with smaller initial
1.5
xr
1.5
vr
1
1.45
15 15.5 16 16.5
0.1
0.5
0.05
0
0
0.5
0
0
10
time (s)
15
20
Fig. 3. Reference signals for wheel velocity and position as described in Eqs. (4) and (5) with t1 1 s, t2 15 s,
t3 16 s, ts 20, vm 0:1 m=s, xd 1:5 m.
2268
errors can reach the switching surface in a shorter time under the proposed SMC, which
makes the control system more robust.
2.4. Equilibrium point of the WIP
At the equilibrium point, the wheel acceleration is zero (x 0), the pendulum angular
velocity and acceleration are zero (y_ 0, y 0), meanwhile the joint friction does not exist
(tf 0), the dynamic Eqs. (1) and (2) become
sin jmp mw g
1
trf r ,
r
mp lg sin y t:
From the above equations, the pendulum equilibrium point is

ye arcsin
r sin jmp mw g rf r
:
mp lg
Remark 4. For the WIP with input coupling, the equilibrium of the pendulum varies with
respecting to the slope angle of the traveling surface and the ground friction. This is
different from WIP system without input coupling, for which the equilibrium of the
pendulum keeps at the upright position, i.e., ye 0.
We give the following explanation based on an intuitive observation. When the WIP
with input coupling is stabilized on a inclined surface (ja0) or traveling at a constant
velocity (fr a0), a torque, denoted as ts , should be provided to overcome the effect of
gravity or the ground friction, meanwhile, the reaction torque ts acts on the pendulum.
The balance of the pendulum can be reached only when the total torque acting on the
pendulum is zero. Thus, the pendulum tilts rightward or leftward from the upright position
such that the torque resulted from the gravity of the pendulum equals to the reaction
torque ts but with the opposite direction.
Considering that our control objective is setpoint control, the WIP nally stops at the
desired setpoint, thus we have fr 0, and ye arcsinr sin jmp mw g=mp lg. To achieve a
zero steady-state error for the pendulum angle, we choose the reference position for the
pendulum as yr ye , that is
yr arcsin
r sin jmp mw
:
mp l
The above yr is applicable only if the system parameters involved are known.
3. Sliding mode controller design
3.1. SMC design for system with unmodeled frictions
The following linear switching surface is proposed
s ce 0,
where c is a constant row vector, and cg is uniformly invertible.
2269
The SMC law is

u
cZ r sgns
,
cg
where sgn is a signum function and

r rm ru r0 ,
10
with rm Zjcgdm j, ru Zjcdu j, and r0 is a positive constant.

Theorem 1. Under the SMC law (9), the WIP system can reach the defined switching surface
(8) in a finite time and maintain on it afterwards. In the sliding mode, the matched
uncertainties will be completely nullified. Furthermore, we have the freedom to choose the
vector c to stabilize the sliding manifold and meanwhile achieve other desirable properties.
Proof. The derivative of s is as
s_ c_e cZ cgu dm cdu :
11
A quadratic Lyapunov function candidate is chosen as

1
V1 s2 :
2
12
Differentiating V1 with respect to time t yields

V_1 ss_ scZ cgu dm cdu :
13
Substituting the control law (9) into Eq. (13), we have

V_1 sr sgnscgdm cdu rr0 jsjo0,
which implies a nite reaching time to the switching surface, s 0, and the reaching time
can be calculated as treach rjs0j=r0 . &
Remark 5. Under the same r0 , the reaching time treach reduces as js0j decreases. As we
stated in Section 2, the absolute value of the initial e1 and e2 are zero by applying the
planned trajectory (4) and (5), which yields a small js0j. Therefore, the switching surface
can be reached in a shorter time.
After reaching the switching surface, the system is in sliding mode and s 0, s_ 0.
Accordingly the equivalent control is derived from s_ 0, as
ueq t
cZ
cdu
dm
:
cg
cg
Dene ed e1,d , e2,d , e3,d , e4,d T as the state vector in the sliding mode, and substitute
the above ueq t into Eq. (3), one obtains the sliding manifold as

cZ
_e d Zed ged
14
deq ,
cg
where

gc
deq I
du :
cg
15
2270
In the sliding manifold, the matched uncertainty dm is completely nullied. Furthermore,

we have the freedom to choose the vector c to stabilize the sliding manifold and meanwhile
achieve other desirable properties, such as robustness, optimality, etc.
3.2. SMC design for system with parameter variations
From the practical point of view, the load of the pendulum mp and slope angle of the
traveling surface j are most likely to vary. The system dynamic model with parameter
uncertainties is expressed as
e_ ge DZe,p ge Dge,pdm u du e,t,
16
where Z and g are known nominal parts, p represents the uncertain parameters, DZ and Dg
are uncertain parts. Dene constants mp,0 , j0 the estimation values of mp and j,
respectively. The known parts are Z Ze,mp,0 ,j0 and g ge,mp,0 ,j0 , respectively. The
unknown parts are DZ Ze,mp ,jZe,mp,0 ,j0 and Dg ge,mp ,jge,mp,0 ,j0 ,
respectively.
The SMC in Eq. (9) is applied with the following switching gain:
r
1
r ru r0 ,
eb m
where
rm Zjcg Dgdm j,

cDgcZ

ru Zjcdu j jcDZj
cg
17
and
r0 40:
Theorem 2. For system with parameter uncertainties, under the SMC law (9) with the
switching gain (17), the WIP system can reach the switching surface (8) in a finite time and
maintain on it afterwards, under the condition jcDgcg1 jo1eb eb 40. In the sliding mode,
the desirable properties stated in Theorem 1 also hold.
Proof. Differentiating the switching surface (8) with respect to time using Eq. (16) one
obtains
_ c_e cZ DZ cg Dgu dm cdu :
st
18
Substituting the SMC law (9) into the above we have

cZ r sgns
_ cZ DZ cg Dg
dm cdu
st
cg
cDg
cDg
r sgns cg Dgdm cdu DZ
cZ:
r sgns
cg
cg
Differentiating the nonnegative quadratic function in Eq. (12) with respect to time t yields
V_ 1 ss_

cDg
cDg
r sgns cg Dgdm cdu DZ
cZ
s r sgns
cg
cg

cDg

jscg Dgdm j jscdu j jscDZj cDg cZs
rrjsj rjsj

cg
cg
2271

cDg
cDg

rjcg Dgdm jjcdu jjcDZj
cZ
jsj r
cg
cg

cDg
rjsj eb rjcg Dgdm jjcdu jjcDZj
cZ :
cg
Substituting the switching gain in Eq. (17) to the above inequality, we have
V_ 1 rr0 jsjo0:
Similarly, we conclude that the switching surface can be reached in a nite time as
treach rjs0j=r0 . In the sliding mode, the equivalent control is derived from s_ 0, which is
cZ DZ
cdu
dm
:
ueq t
cg Dg
cg Dg
Substituting the above ueq(t) into Eq. (16), one obtains the sliding manifold as Eq. (14) with

g Dgc
cZ
deq I
du DZu g
:
19
cg Dg
cg
In the sliding manifold, the matched uncertainty dm is completely nullied. Sliding motion
is determined by the vector c. &
4. Switching surface design
Under the proposed SMC in Eq. (9), the WIP system can reach the switching surface in a
nite time. When the system is in sliding mode, from Eq. (14), we can see that the sliding
motion is directly determined by the vector c. In [1012], the stabilization of the sliding
manifold is achieved by choosing the switching surface coefcients according to several
established constraints. However, since the vector c affects the system performance in a
complicated manner, it is hard to predict the system responses from the information of c,
which is the main drawback of the existing switching surface designs. Another major
drawback or difculty in the switching surface design is the nonafne structure of the
sliding manifold in the coefcients c, as shown in Eq. (14). To avoid the drawbacks, in this
work, the switching surface coefcients are determined indirectly, by transforming the
coefcients determination problem into a nominal linear controller design. Hence the
switching surface design becomes simple, systematic, and furthermore provides one extra
degree of freedom in control. Feedback gains for the nominal controller can be determined
through various systematic linear control design methods, which makes the system
responses predictable. In the transformation, relations are established between the
switching surface coefcients and the feedback gains. With the obtained feedback gains,
the switching surface coefcients can be determined.
4.1. Problem transformation

We linearize the sliding manifold (14) around the desired equilibrium point by assuming
e3,d 0, sin e3,d e3,d and e24,d 0. The obtained linearized sliding manifold is
e_ d Aed g0 u0 deq,0 ,
20
2272
where
2
0 1
60 0
6
A6
40 0
0 0
0
a23
0
a43
07
7
7,
15
0
7
6
6 g1,0 7
7
g0 6
6 0 7,
5
4
g2,0

gc
deq,0 I 0 du ,
cg0
with
a23
b0 mp lg cos yr
,
acb20
a43
c
b0
,
2
racb0 acb20
b0 mp l cosj yr ,
g1,0
amp lg cos yr
,
acb20
g2,0
b0
a
,

2
racb0 acb20
and the system control input is as

c1
c2 a23 c4 a43
c3
u0
e2,d
e3,d
e4,d :
c2 g1,0 c4 g2,0
c2 g1,0 c4 g2,0
c2 g1,0 c4 g2,0
21
When the WIP system is in the sliding mode, we have s c1 e1,d c2 e2,d c3 e3,d
c4 e4,d 0, thus,
c2
c3
c4
e1,d e2,d e3,d e4,d :
22
c1
c1
c1
Let
c1
k1
k2 c2 ,
c2 g1,0 c4 g2,0
c1
23
c2 a23 c4 a43
k1
k3 c3 ,
c2 g1,0 c4 g2,0
c1
24
c3
k1
k4 c4 ,
c2 g1,0 c4 g2,0
c1
25
the system control input (21) can be rewritten as

k1
k1
k1
u0 k2 c2 e2,d k3 c3 e3,d k4 c4 e4,d
c1
c1
c1

c2
c3
c4
k1
e2,d e3,d e4,d k2 e2,d k3 e3,d k4 e4,d
c1
c1
c1
k1 e1,d k2 e2,d k3 e3,d k4 e4,d :
26
To stabilize the nominal linear system, i.e., the linearized sliding manifold (20), various
linear controller design methods could be applied to obtain the feedback gains
k k1 , k2 , k3 , k4 , and c1 2c4 can be solved from relations (23)(25).
Remark 6. The four coefcients c1 2c4 are constrained by three Eqs. (23)(25), therefore
there are innumerable solutions of c for a given k. However, from Eqs. (23)(25), it can be
2273
concluded that the ratios between c1 2c4 are xed. In this work, we choose c1 be the free
parameter, then c2 2c4 can be decided from Eqs. (23)(25) once c1 is set.
4.2. Nominal linear controller design
Two alternative linear control design methods are introduced in this work. For WIP
system to travel in a safe environment, a LQR based optimal linear controller is adopted to
achieve an stable and optimal sliding manifold. For the WIP system to travel in a severe
environment with various uncertainties, robustness of the system is the main concern, thus
a LMI based robust design is employed to address unmatched uncertainties.
4.2.1. LQR based optimal design
LQR is a widely used method owing to its good properties in terms of stability and
optimality. An LQR problem is to minimize an index
Z
1 1 T
JLQR
e Qe uT Ru dt,
27
2 0
with QZ0 and R40.
The solution for the optimal control gain is as
k R1 gT0 P1 ,
28
where P1 is the solution of the following Riccati equation

P1 A AT P1 P1 g0 R1 gT0 P1 Q 0:
In this work, we choose the weighting matrix Q in diagonal form, i.e., Q diagfq1 ,
q2 ,q3 ,q4 g, and R a scalar, where qi and R are the weighting factors for ei (i 1, 2, 3, 4) and
u, respectively.
The relative values of qi represent the relative weighting among ei. If q1 is bigger than q2,
there is higher penalty on error e1 than e2 and control tries to make smaller e21 than e22 , and
vice versa. For control of the WIP system, the main objective is the convergence of e1 and
e3, thus it is nature to select q1 and q3 to be relatively larger than q2 and q4. Furthermore,
considering that the balancing of the pendulum is more important than the tracking
performance of the wheel, it is reasonable to select q3 and q4 to be relatively larger than
q1 and q2, respectively.
The relative value of the weighting matrix Q and R expresses the relative importance of
keeping e and u near zero. If we place more importance on the convergence of e, then we
can select Q to be relatively large to R, and so forth. Although we are interested in
minimizing JLQR in Eq. (27), the actual value of JLQR is usually not of interest, which also
means that we can set either Q or R to be xed for the convenience of parameter tuning
because it is their relative weight that is important. In this work, since Q is matrix and R is
a scalar, it is better to x Q and tune R to achieve desired performance. A smaller R results
in a larger feedback gain and faster convergence of e, however, a larger magnitude of u.
Thus, the selection of R should achieve a compromise between these effects.
The control system based on LQR design not only shows optimality, but also
robustness. Applying the LQR based linear controller, the phase margin of the closed-loop
system is at least 60 degrees and the gain margin is from 1/2 to innity.
2274
4.2.2. LMI based robust design

Let k wP, we have u0 wPe. The nominal linear system (20) becomes
e_ d Aed g0 wPed deq,0 :
Dene a Lyapunov function
29
V2 eTd Ped ,
differentiating V2 with respect to t yields
V_ 2 e_ Td Ped eTd P_e d

Aed g0 wPed deq,0 T Ped eTd PfAed g0 wPed deq,0 g
V_ 2 eTd Ag0 wPT P PAg0 wPed dTeq,0 Ped eTd Pdeq,0 :
30
31
For deq,0 0, to make V_ o0, we have the following sufcient condition:

Ag0 wPT P PAg0 wP mIo0,
32
with mZ0.
Assume the unmatched uncertainties deq,0 is bounded by b1 Jed J b2 with b1 ,b2 40, we
have
dTeq,0 Ped eTd Pdeq,0 r2lmax Pb1 Jed J22 b2 Jed J2 ,
33
thus
V_ 2 omJed J22 2lmax Pb1 Jed J22 b2 Jed J2 :
For system with only vanishing unmatched uncertainties, i.e., b1 a0 and b2 0, we have
V_ o0 if b1 om=2lmax P. It can be concluded that the desired equilibrium ed 0 is locally
asymptotically stable. For system with both vanishing and nonvanishing unmatched
uncertainties, i.e., b1 a0 and b2 a0, we have V_ 2 is negative outside the set fJed Jr
2b2 lmax P=m2b1 lmax Pg, under the condition that b1 om=2lmax P. We can conclude
that Jed J is ultimately bounded by
Jed Jr
2b2 lmax P
m2b1 lmax P
b2
b
2lmax P 1
Based on the above analysis, we seek for solutions of w and P which can maximize
m=2lmax P such that the system could be robust against unmatched uncertainties,
meanwhile the ultimate bound of Jed J can be minimized.
Pre and post multiplying Eq. (32) by P1 , and letting P P1 , we have
2
PAT APwT gT0 g0 w mP o0,

which can be casted in the LMI form as
8
m
>
>
> max2l P
>
max
>
>
<"
p #
PAT APwT gT0 g0 w
mP
p
o0 :
>
>
m
P
I
>
>
>
>
:
Po0
The above LMIs can be solved numerically.
34
35
2275
5. Numerical validations
For simulation, fr is modeled as a combination of viscous friction and Coulomb friction,
_ where fv 0.2 and fc 0.3. Similarly, tf is modeled as
that is, fr fv x_ fc sgnx,
_ where tv 0:2 and tc 0:3. Both frictions are vanishing terms and
tf tv y_ tc sgny,
assumed to be unknown. From the dynamic Eq. (3), it is evident that tf is the matched
uncertainty to the control input while fr is the unmatched uncertainty. Initial states for the
WIP system are x 0,0,0:1,0T .
Case 1. In this case, a linear controller based on LQR method [29], is applied to the WIP
system with and without the joint friction tf . The WIP travels on a at surface. Choose fq1 , q2 ,
q3 , q4 g f50, 0:1, 500, 1g, R 0.8. We obtain the feedback gains as k 7:9057,
10:7948, 29:9739,3:1183T . The results are shown in Fig. 4. For WIP system without
the joint friction, the LQR based linear controller shows effectiveness that the wheel reaches the
desired setpoint smoothly with a small overshoot, and the pendulum angle stays around zero.
However, the LQR based linear controller cannot function well when the joint friction exists in
the WIP system. The pendulum and the wheel keep vibrating around the desired positions,
which does not meet the desired performance.
Case 2. In this case, we consider only the joint friction tf exists in the WIP system, which
is a matched uncertainty. The WIP travels on a at surface. SMC is applied with
parameters designed as the following. Refer to Eq. (7), the reference position for the
pendulum is yr 0 since j 0, the switching gain is r 0:1 jcgj0:3 0:2jx4 j. Feedback
gains for the nominal controller u0 are obtained based on LQR method.
(a) To compare the WIP responses under the SMC and the LQR based linear controller,
the weighting matrices Q and R are chosen the same as in Case 1. The feedback gains
are obtained as k 7:9057, 10:7948, 29:9739, 3:1183. Set c1 1, solving Eqs.
(23)(25) yields c2 , c3 , c4 1:330530, 3:454665, 0:2738175. The results are shown in
Fig. 5. The WIP reaches the desired setpoint smoothly and stays still afterwards, the
pendulum is balanced at ye 0 and the control signal shows switching behavior. The
switching surface is reached at t 0.87 s, which is a fairly short time.
Comparing the results in Case 2(a) with the results in Case 1 when tf 0, the
performances are almost the same in terms of tracking error proles and control
1.6
Wheel Position x (m)
1.2
1
0.8
f 0
0.6
reference
0.4
0.2
0
10
20
time(s)
30
3
=0
f
0.08
f 0
0.06
Control Signal u (Nm)
Pendulum Tilting Angular (rad)
0.1
1.4
0.04
0.02
0
0.02
0.04
=0
f
2.5
f 0
2
1.5
1
0.5
0
10
20
time(s)
30
0.5
10
20
30
time(s)
Fig. 4. Case 1: time responses of x, y and u under linear controller based on LQR. WIP travels on a at surface.
System with and without the joint friction tf are considered.
1.4
1.2
1
0.8
x1
0.6
xr
0.4
0.2
0
0.1
0.8
0.08
0.6
0.06
0.04
0.02
0
10
20
time (s)
30
0.3
0.4
0.2
0
0.2
0.2
0.1
0
0.4
0.02
0.04
0.4
Sliding Manifold
1.6
2276
0.1
0.6
0
10
20
time (s)
30
10
20
time (s)
30
10
20
30
time (s)
Fig. 5. Case 2 (a): time responses of x, y, u and s under SMC. WIP is considered traveling on a at surface. Only
_ exists in the system. R 0.8 is used in SMC design.
matched uncertainty, the joint friction tf 0:2y_ 0:3 sgny,
proles. However, in Case 1 the control system is directly designed by LQR, whereas in
Case 2(a), the control system is rst designed by SMC in order to eliminate the effect
caused by the matched uncertainty, and then the switching surface is designed by LQR
for the sliding manifold. Thus, our new SMC approach achieves both robust and
optimal properties.
(b) To illustrate our discussion in Section 4 that the selection of the weighting matrices
directly affects the system performance, we select the weighting matrix Q to be the
same as in Case 2(a), while the weighting factor for the control input to be R 0.08,
which is smaller than the one used in Case 2(a). The obtained feedback gains are
k 25:0000, 33:7972, 90:3320, 8:6977, which are larger than in Case 2 (a). Let
c1 1, solving Eqs. (23)(25) yields c2 , c3 , c4 1:307674, 3:338392, 0:200306. The
results are shown in Fig. 6. The WIP system is stabilized and reached the desired
setpoint. We can see the switching amplitude of the control signal in Case 2(b) is much
larger than the one in Case 2(a), which is due to the higher feedback gains used for the
nominal linear controller in Case 2(b). The results are consistent with the discussions in
Section 4.
Remark 7. It is noticeable that the vector c used in Case 2(b) is quite close to the one in
Case 2(a). In other words, a minor change of the vector c might lead to large changes in the
system responses and the changes are unpredictable, which indicates the difculty in tuning
the vector c directly to achieve desired responses and shows the drawback of the existing
switching surface designs. With our proposed design method, the switching surface
coefcients are determined indirectly. The sliding manifold is determined by the feedback
gains of the nominal linear controller, which can be tuned in a systematic way. For
instance, we can choose more of control penalty in R so as to prevent overlarge control
signals. The advantage of our proposed method is immediately obvious.
Case 3. In this case, we consider both frictions, tf and fr, exist in the WIP system, i.e., the
system is in the presence of both matched and unmatched uncertainties. The WIP travels
on a at surface. SMC is applied with r 0:1 jcgj0:3 0:2jx4 j jcc2 bc4 =
acb2 j 0:3 0:2jx2 j, and all other parameters the same as in Case 2. The results are
shown in Fig. 7. The WIP reaches the desired setpoint and the pendulum is nally balanced
1.2
x1
xr
1
0.8
0.6
0.4
0.2
0
0
10
20
time(s)
0.1
1.5
0.08
0.06
0.04
0.02
0
0.3
0.5
0
0.5
1.5
0
10
20
time(s)
0.1
0.1
0
30
0.2
0.02
0.04
30
0.4
Sliding Manifold
1.4
1.6
2277
10
20
time(s)
30
10
20
time(s)
30
Fig. 6. Case 2 (b): time responses of x, y, u and s under SMC. WIP is considered traveling on a at surface. Only
_ exists in the system. R 0.08 is used in SMC design.
1.2
1
0.8
x1
0.6
xr
0.4
0.2
0
0
10
20
time (s)
30
0.1
0.8
0.08
0.6
0.06
0.04
0.02
0
0.02
0.4
0.3
Sliding Manifold
1.4
1.6
0.4
0.2
0
0.2
20
time (s)
30
0.6
0.1
0
0.4
10
0.2
0.1
0
10
20
time (s)
30
10
20
30
time (s)
Fig. 7. Case 3: time responses of x, y, u and s under SMC. WIP is considered traveling on a at surface. Both
_ and unmatched uncertainty, the ground friction
_ exist in the system.
fr 0:2x_ 0:3 sgnx,
at the upright position, i.e., ye 0, which indicates that the proposed SMC is also robust to
unmatched uncertainties.
Case 4. In this case, the WIP traveling on an inclined surface is considered. The slope
angle is known as j p=15 rad. Both frictions tf and fr exist in the system. SMC is applied
with parameters designed as the following. Refer to Eq. (7), the reference position for the
pendulum is yr 0:2547 rad. Feedback gains k 7:9057,10:7535,30:0154,3:1275
for the nominal controller u0 are obtained based on LQR method, where Q and R are
chosen the same as in Case 1. Next, let c1 1, solving Eqs. (23)(25) yields
c2 ,c3 ,c4 1:327784, 3:470293, 0:283022. The switching gain is the same as in Case 3.
The results are shown in Fig. 8, the WIP reaches the desired setpoint, while the pendulum
is balanced at the new equilibrium point ye yr 0:2547 rad. The simulation results are
consistent with the theoretical analysis in Section 2.4.
Case5. In this case, the WIP travels on the same surface as in Case 4. However, the slope
angle is assumed to be unknown to the designer, thus j0 0 and yr 0 are used in the
controller design. Both frictions tf and fr exist in the system. SMC is applied. Feedback
gains k 7:4802, 11:2445, 26:9865, 5:5473 for the nominal controller u0 are
obtained based on LMI method. Next, let c1 1, solving Eqs. (23)(25) yields
c2 , c3 , c4 1:241766, 2:113164, 0:1890710. The results are shown in Fig. 9. The
1
x1
0.5
xr
0
0.5
10
20
0.1
0
0.3
0.25
0.2
0.15
0.1
30
1.5
0.35
Sliding Mani fold
1.5
2278
0.5
0.1
0.2
0.3
0.4
0.5
0.5
0
time(s)
10
20
time(s)
30
10
20
0.6
30
time(s)
10
20
30
time(s)
Fig. 8. Case 4: time responses of x, y, u and s under SMC. WIP is considered traveling on an inclined surface with
_ and unmatched
known slope angle j p=15. Both matched uncertainty, the joint friction tf 0:2y_ 0:3 sgny,
uncertainty, the ground friction fr 0:2x_ 0:3 sgnx,
0.5
x1
xr
0.5
0
10
20
time(s)
30
0.3
0.25
0.2
0.15
0.1
0.05
0.3
0.25
4
Sliding Manifold
0.35
1.5
2
0
2
10
20
time(s)
30
0.1
0.05
0
0.2
0.15
0.05
0
10
20
time(s)
30
10
20
30
time(s)
Fig. 9. Case 5: time responses of x, y, u and s under SMC. WIP is considered traveling on an inclined surface with
_ and unmatched
unknown slope angle. Both matched uncertainty, the joint friction tf 0:2y_ 0:3 sgny,
uncertainty, the ground friction fr 0:2x_ 0:3 sgnx
pendulum is balanced at the new balanced position ye , which is around y 0:2547 rad,
thus steady state error for e3 exist as e3,s ye yr a0. From s ce 0, steady-state error
for e1 also exist as e1,s c3 e3,s =c1 0:5382 m, which meets the simulation results.
Comparing the results in Case 4 and Case 5, we can see the necessity of adjusting the
reference position of pendulum when the WIP travels on an inclined surface. yr cannot be
computed according to Eq. (7) if system parameters involved are unknown. When task
repeats, we obtain yr ye , which can be incorporated in the controller design to eliminate
steady-state errors.
6. Implementation and experiment results
In simulations, an ideal model of the WIP is used. To stabilize the WIP system in
absence of uncertainties, the feedback gains for the nominal linear controller can be chosen
in a wide range as long as Ag0 k is Hurwitz. However, considering the existence of
mismatch between the real-time system model and the mathematical model (1) and (2), the
feedback gains obtained from simulations may not function well on the real-time platform,
thus need to be adjusted through experimental testings on the WIP prototype.
2279
6.1. Regulation task

For implementation, rst, we consider a simple regulation task that is to balance the
WIP at the original position on a at surface, i.e., xr 0, vr 0, and j 0. In order to nd
appropriate feedback gains, rst a linear controller is applied with the feedback gains
obtained from simulations. However, strong vibrations are observed, which can be
explained as the following. For systems having backlash in the driving mechanism, large
feedback gains could easily incur vibrations [9]. In our WIP system, the backlash is
produced by the gearbox. To reduce or avoid vibrations, feedback gains for the velocity
terms, k2 and k4, should be minimized [9]. From experimental testing, we observe that the
system vibration reduces signicantly as k2 and k4 decrease. It is also found that large
feedback gains are necessarily needed for the position terms, hence, k1 and k3 should be
kept at the original designed level. If k3 is too small, the pendulum could easily fall down
because the torque generated by the motor is not large enough to overcome the effect of
the gravity. If k1 is too small, the position control of the wheels fails that a steady state
error exists in the response of the wheel position. After several trials, the feedback gains are
adjusted to k 7:9057, 0:5, 29:9739, 1:8.
For implementing the SMC, the obtained feedback gains are used for the nominal linear
controller u0. Let c1 1, solving Eqs. (23)(25) yields c2 , c3 , c4 0:026051, 3:627349,
0:092766. Fig. 10 shows the experimental results for the WIP under the SMC and the
linear controller. When the linear controller is applied, the WIP becomes unstable in 10 s.
By applying the SMC, the WIP is stabilized. The wheels stay around the original place and
the pendulum is balanced around y 0, which veries the effectiveness of the SMC in
handling system uncertainties. The value of the dened sliding variable s switches
around zero.
0.5
SMC
LC
0
SMC
LC
0.5
1
(rad)
x (m)
0.5
10
15
20
25
0.5
30
10
30
20
30
0.1
0.05
u (Nm)
10
10
20
20
time (s)
time (s)
SMC
LC
0
10
15
time (s)
20
25
0
0.05
30
0.1
10
time (s)
Fig. 10. Experimental testing results for regulation task: time responses of x, y, u and s under SMC and linear
controller. The WIP is placed on at surface.
2280
0.2
(rad)
x (m)
0.1
1
x1
xr
0.1
0.2
1
0
10
15
time (s)
20
25
10
15
time (s)
20
25
10
15
time (s)
20
25
0.2
0.1
u (N m)
0
0.1
0.2
0
10
15
time (s)
20
25
Fig. 11. Experimental testing results for setpoint task: time responses of x, y, u and s under SMC. The WIP is
placed on a at surface. The reference trajectory (5) is applied.
6.2. Reaching a setpoint

We consider the WIP travels on a at surface, i.e., j 0. The reference trajectory for the
WIP is the same as in simulations. SMC is applied and the sliding surface coefcients are
chosen the same as for the regulation task. Experiment results are shown in Fig. 11.
The WIP reached the desired setpoint x 1.5 m and stays there afterwards. The pendulum
is stabilized around the upright position. SMC shows effectiveness for setpoint control of
the WIP system.
7. Conclusion
In this work, control of an underactuated WIP system with input coupling is studied. A
novel design of SMC is presented. First, a linear switching surface and the SMC is
introduced. The system reaches the switching surface in a nite time under the proposed
SMC. Next, after the system reaches the switching surface, the switching surface design is
discussed, which mainly focuses on choosing the switching surface coefcients to stabilize
the sliding manifold. To avoid the complexity on tuning the coefcients directly, we
transform the problem into a simple nominal linear controller design problem, which not
only simplies the tuning process, but also provides one extra degree of freedom in control.
By utilizing the extra degree of freedom, optimal and robust linear control techniques are
incorporated in the SMC design. Intensive simulations and experiment testings are
conducted to verify the effectiveness of the proposed SMC and satisfactory results are
achieved. Thanks to the easiness in tuning the feedback gains for the nominal linear
2281
controller, the proposed SMC shows realizability and superiority regarding the
implementation on real-time platforms of underactuated mechanical systems.
References
[1] O.-S. Reza, Nonlinear Control of Underactuated Mechanical Systems with Application to Robotics and
Aerospace Vehicles, MIT PhD Thesis, 2001.
[2] I. Fantoni, R. Lozano, Nonlinear Control for Underactuated Mechanical Systems, 2002.
[3] F. Grasser, A. DArrigo, S. Colombi, A.C. Rufer, JOE: a mobile, inverted pendulum, IEEE Transaction on
Industrial Electronics 49 (1) (2002) 107114.
[4] J. Solis, A. Takanishi, Development of a wheeled inverted pendulum robot and a pilot experiment with
master students, in: Proceedings of the 7th International Symposium on Mechatronics and its Applications
(ISMA10), Sharjah, UAE, 2010, pp. 16.
[5] J. Solis, R. Nakadate, T. Yamamoto, A. Takanishi, Introduction of mechatronics to undergraduate students
based on robotic platforms for education purposes, in: Proceedings of the 18th IEEE International
Symposium on Robot and Human Interactive Communication Toyama, Japan, 2009, pp. 693698.
[6] P. Oryschuk, A. Salerno, A.M. Al-Husseini, J. Angeles, Experimental validation of an underactuated twowheeled mobile robot, IEEE/ASME Transactions on Mechatronics 14 (2) (2009) 252257.
[7] C.-H. Chiu, Y.-W. Lin, C.-H. Lin, Real-time control of a wheeled inverted pendulum based on an intelligent
model free controller, Mechatronics 21 (2011) 523533.
[8] C.-H. Huang, W.-J. Wang, C.-H. Chiu, Design and implementation of fuzzy control on a two-wheel inverted
pendulum, IEEE Transactions on Industrial Electronics 58 (7) (2011) 29883001.
[9] T. Takei, R. Imamura, S. Yuta, Baggage transportation and navigation by a wheeled inverted pendulum
mobile robot, IEEE Transactions on Industrial Electronics 56 (10) (2009) 39853994.
[10] M.-S. Park, D. Chwa, Swing-up and stabilization control of inverted-pendulum systems via coupled slidingmode control Method, IEEE Transactions on Industrial Electronics 56 (9) (2009) 35413555.
[11] J. Huang, Z.-H. Guan, T. Matsuno, T. Fukuda, K. Sekiyama, Sliding-mode velocity control of mobilewheeled inverted-pendulum systems, IEEE Transactions on Robotics 26 (4) (2010) 750758.
[12] M.-S. Park, D. Chwa, Orbital stabilization of inverted-pendulum systems via coupled sliding-mode control,
IEEE Transactions on Industrial Electronics 56 (9) (2009) 35563570.
[13] H. Ashrauon, R.S. Erwin, Sliding mode control of underactuated multibody systems and its application to
shape change control, International Journal of Control 81 (12) (2008) 18491858.
[14] R. Xu, U. Ozguner, Sliding mode control of a class of underactuated systems, Automatica 44 (2008) 233241.
[15] S. Kurode, P. Trivedi, B. Bandyopadhyay, P.S. Gandhi, Second order sliding mode control for a class of
underactuated systems, in: Proceedings of the 12th IEEE International Workshop on Variable Structure
Systems (VSS12), 2012, pp. 458462.
[16] S. Riachy, Y. Orlov, T. Floquet, R. Santiesteban, J. Richard, Second order sliding mode control of
underactuated mechanical systems I: local stabilization with application to an inverted pendulum,
International Journal of Robust Nonlinear Control 18 (4/5) (2007) 529543.
[17] R. Santiesteban, T. Floquet, Y. Orlov, S. Riachy, J. Richard, Second-order sliding mode control of
underactuated mechanical systems II: orbital stabilization of an inverted pendulum with application to swing
up/balancing control, International Journal of Robust Nonlinear Control 18 (4/5) (2007) 544556.
[18] R. Iriarte, L. T. Aguilar, L. Fridman, Second order sliding mode tracking controller for inertia wheel
pendulum, Journal of the Franklin Institute, 2012, http://dx.doi.org/10.1016/j.jfranklin.2012.10.013.
[19] M. Lopez-Martinez, J.A. Acosta, J.M. Cano, Non-linear sliding mode surfaces for a class of underactuated
mechanical systems, IET Control Theory and Applications 4 (10) (2010) 21952204.
[20] J.-X. Xu, Z.-Q. Guo, T.H. Lee, Synthesized design of a fuzzy logic controller for an underactuated unicycle,
Fuzzy Sets and Systems 207 (2012) 7793.
[21] V. Azhmyakov, On the set-valued approach to optimal control of sliding mode processes, Journal of the
Franklin Institute-Engineering and Applied Mathematics 349 (4) (2012) 13231336.
[22] M. Basin, P. Rodriguez-Ramirez, Sliding mode controller design for linear systems with unmeasured states,
Journal of the Franklin Institute-Engineering and Applied Mathematics 349 (4) (2012) 13371349.
[23] M. Basin, P. Rodriguez-Ramirez, A. Ferrara, D. Calderon-Alvarez, Sliding mode optimal control for linear
systems, Journal of the Franklin Institute-Engineering and Applied Mathematics 349 (4) (2012) 13501363.
2282
[24] B. Mirkin, P.-O. Gutman, Y. Shtessel, Coordinated decentralized sliding mode MRAC with control cost
optimization for a class of nonlinear systems, Journal of the Franklin Institute-Engineering and Applied
Mathematics 349 (4) (2012) 13641379.
[25] A. Polyakov, Minimization of disturbances effects in time delay predictor-based sliding mode control
systems, Journal of the Franklin Institute-Engineering and Applied Mathematics 349 (4) (2012) 13801396.
[26] T.R. Oliveira, A.J. Peixoto, L. Hsu, Global real-time optimization by output-feedback extremum-seeking
control with sliding modes, Journal of the Franklin Institute-Engineering and Applied Mathematics 349 (4)
(2012) 13971415.
[27] A.F. de Loza, M. Jimenez-Lizarraga, L. Fridman, Robust output nash strategies based on sliding mode
observation in a two-player differential game, Journal of the Franklin Institute-Engineering and Applied
Mathematics 349 (4) (2012) 14161429.
[28] J.-X. Xu, A quasi-optimal sliding mode control scheme based on control Lyapunov function, Journal of the
Franklin Institute-Engineering and Applied Mathematics 349 (4) (2012) 14451458.
[29] S.Y. Seo, S.H. Kim, S.-H. Lee, S.H. Han, H.S. Kim, Simulation of attitude control of a wheeled inverted
pendulum, in: Proceedings of the International Conference on Control, Automation and Systems, Korea,
2007, pp. 22642269.

Inverted Pendulum

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Inverted Pendulum

Încărcat de

Drepturi de autor:

Formate disponibile

Available online at www.sciencedirect.

Journal of the Franklin Institute 351 (2014) 22612282

Design and implementation of a new sliding mode

Throughout this paper, a function F l1 ,l2 , . . . ,ln will be written as F, where

Fig. 1. Prototype of the WIP.

Fig. 2. Model of the WIP system.

where a mw mp Iw =r2 , b mp l cosx3 j and c Ip mp l 2 .

and xr ts xd , where xd is the setpoint.

Ref for Wheel Position (m) / Vel (m/s)

mp lg sin y t:

From the above equations, the pendulum equilibrium point is

The SMC law is

where sgn is a signum function and

with rm Zjcgdm j, ru Zjcdu j, and r0 is a positive constant.

A quadratic Lyapunov function candidate is chosen as

Differentiating V1 with respect to time t yields

Substituting the control law (9) into Eq. (13), we have

In the sliding manifold, the matched uncertainty dm is completely nullied. Furthermore,

Substituting the SMC law (9) into the above we have

4.1. Problem transformation

and the system control input is as

the system control input (21) can be rewritten as

where P1 is the solution of the following Riccati equation

4.2.2. LMI based robust design

differentiating V2 with respect to t yields

V_ 2 e_ Td Ped eTd P_e d

For deq,0 0, to make V_ o0, we have the following sufcient condition:

PAT APwT gT0 g0 w mP o0,

Wheel Position x (m)

Control Signal u (Nm)

Pendulum Tilting Angular (rad)

Wheel Position x (m)

Pendulum Tilting Angular (rad)

Control Signal u (Nm)

Wheel Position x (m)

Control Signal u (Nm)

Pendulum Tilting Angular (rad)

Wheel Position x (m)

Control Signal u (Nm)

Pendulum Tilting Angular (rad)

Sliding Mani fold

Pendulum Tilting Angular (rad)

Wheel Position x (m)

Control Signal u (Nm)

Pendulum Tilting Angular (rad)

Wheel Position x (m)

6.1. Regulation task

6.2. Reaching a setpoint

S-ar putea să vă placă și

where sgn is a signum function and