2009 IEEE International Conference on Robotics and Automation

Kobe International Conference Center

Kobe, Japan, May 12-17, 2009

Robust Velocity Sliding Mode Control of Mobile Wheeled Inverted

Pendulum Systems
Jian Huang, Member, IEEE, Hongwei Wang, Takayuki Matsuno, Toshio Fukuda, Fellow, IEEE, and
Kousuke Sekiyama

AbstractThere has been an increasing interest in a kind of

underactuated mechanical systems, mobile wheeled inverted
pendulum (MWIP) models, which are widely used in the field of
autonomous robotics and intelligent vehicles. Robust velocity
tracking problem of MWIP systems is investigated in this study.
In the velocity control problem, model uncertainties accompany
uncertain equilibriums, which make the controller design
become more difficult. A sliding mode control (SMC) method
based on a novel sliding surface is proposed for the systems,
which are capable of handling both parameter uncertainties and
external disturbances. By assuming the specially designed
sliding surface, the proposed SMC controller is capable of
eliminating the steady velocity tracking error. The asymptotical
stability of the closed-loop system is achieved through selecting
sliding surface parameters in terms of some rules. The
effectiveness of the proposed methods is finally confirmed by
numerical simulations.

I. INTRODUCTION

ECENTLY many investigations have been devoted to

problems of controlling mobile wheeled inverted
pendulum (MWIP) models, which are widely taken into
applications in the field of autonomous robotics and
intelligent vehicles [1-10]. The MWIP models are not only of
theoretical interest but also of practical interest. Many
practical systems have been implemented based on MWIP
models such as JOE[5], Nbot[6], Legway[7], B2[8],
Segway[9] and so on. Among these applications, the Segway
PT has been a popular personal transporter since invented in
2001. Such systems are characterized by the ability to balance
on its two wheels and spin on the spot. This additional
maneuverability allows easy navigation on various terrains,
turn sharp corners and traverse small steps or curbs. In
addition, the compact structure design allows drivers to
access most places that can only be accessed by walkers in the
Manuscript received September 15, 2008. The first author thanks the
support of the National Natural Science Foundation of China under Grants
60603006 and the Hi-tech Research and Development Program of China
under Grant 2008AA04Z207. Jian Huang and Hongwei Wang are with the
Department of Control Science & Engineering, Huazhong University of
Science and Technology, Wuhan, Hubei, 430074, P.R. China. (e-mail:
huang@robo.mein.nagoya-u.ac.jp, hwwang@mail.hust.edu.cn)
Takayuki Matsuno is with the Department of Intelligent System Design
Engineering, Toyama Prefectural University, Imizu City 939-0398, Japan
(e-mail:matsuno@pu-toyama.ac.jp).
Toshio Fukuda and Kosuke Sekiyama are with the Department of
Micro-Nano Systems Engineering, Nagoya University, Furo-cho,
Chikusa-ku, Nagoya 464-8603, Japan (phone: +81-52-789-4481; fax:
+81-52-789-3115;
e-mail:
fukuda@mein.nagoya-u.ac.jp,
sekiyama@robo.mein.nagoya-u.ac.jp).

978-1-4244-2789-5/09/$25.00 2009 IEEE

past. Moreover, people can drive such vehicles to travel short

distances in a small area instead of using cars or buggies
which are more pollutive.
From the point of view of theories, the MWIP models have
attracted much attention in the field of control theory and
engineering because they are nonlinear and underactuated
with inherent unstable dynamics. Many control techniques
have been studied in the past decades for the control of
benchmark underactuated systems such as the inverted
pendulum, the acrobot and the rotating pendulum. A control
technique comprised of swing-up and balancing controller
was designed using partial feedback linearization to balance
the pendulum [11]. Other methods, like passivity-based,
adaptive and controlled Lagrangians techniques, have also
found their way into the control of underactuated mechanical
systems [12-14]. Nevertheless, most of these methods are
based on exact mathematical models and lack of robustness to
model errors and external disturbances. In the case of MWIP
models, although they are essentially nonlinear and
underactuated, most of the previous work used linear [2, 3] or
feedback linearization methods [1, 10] in the modeling and
control of them. It is well known that linear feedback
controller of a nonlinear system yields a small region of
attraction and lacks robustness to parameter uncertainty. On
the other hand, feedback linearization involves the exact
cancellation of nonlinearities. Consequently, it relies on a
rather precise description of nonlinear functions. H method
has also been applied to control the B2 vehicle, although it is
still based on a reference linear model and only the balancing
problem was investigated [15].
The sliding-mode control (SMC) might be a comparatively
appropriate approach to deal with uncertain MWIP systems
because SMC is less sensitive to the parameter variations and
noise disturbances. Whereas, it is not easy to design SMC
controllers for nonlinear underactuated systems, whose
control problems have been proved to be challenging because
the techniques of fully actuated systems cannot be used
directly. Wang proposed a hierarchical SMC method for a
class of second-order underactuated systems [16].
Nevertheless the results are still in dispute [17]. Ashrafiuon
proposed an SMC approach for underactuated multibody
systems [18]. This approach was further applied into a robust
tracking control problem of biped robots [19]. In their study,
external disturbances are paid more attention to than
parameter uncertainty. As for special cases of underactuated
systems, Lee and Su applied the SMC methods to the robust

2983

M 11&&w + (M 12 cos(1 + )) &&1 =

~
Dw&w + D1 &1 &w + M 12&12 sin (1 + ) M

&&
&&
(M 12 cos(1 + )) w + M 221 =
(1)
D & & + G sin
1 1
w
1
1

control of the Acrobot and the underactuated manipulator

respectively [20, 21].

where parameters M 11 , M 12 , M 22 and G1 satisfy

M 11 = (m1 + mw ) r 2 + I w

M 12 = m1lr

2
M 22 = m1l + I1
G = m gl
1
~1
M = (m1 + mw )rg sin .

Fig. 1. The mobile wheeled inverted pendulum system model

In this paper, a SMC approach based on a novel sliding

surface is proposed to realize the velocity tracking problem of
a mobile wheeled inverted pendulum. Both model
uncertainties and external disturbances are taken into account
in the controller design. The model formulation and
equilibrium analysis are discussed in Section II. The
procedure of the design for the controller is discussed in
Section III. In Section IV, the effectiveness of the proposed
methods is proved through numerical simulations.
II. SYSTEM FORMULATION
A. Dynamic model
The MWIP system is modeled as a one dimensional inverse
pendulum which rotates about the wheels axles. Hence, the
bodys motion on a plane is determined by the inclination and
translational motion. Fig. 1 shows the structure of an MWIP
system, where w and 1 are the wheels rotation angle and
the inclination angle of the body, respectively. We suppose
that the system moves on a slope with constant but uncertain
angle , which is hard to be measured by sensors. To
describe the parameters of the MWIP system, some notations
should be clarified first (see also Fig. 1):
m1 , mw - masses of the body and the wheel.

I1 , I w - moments of inertia of the body and the wheel.

l - length between the wheel axle and the center of gravity
of the body.
r - radius of the wheel.
D1 - viscous resistance in the driving system.

Dw - viscous resistance of the ground.

Lagrange's motion equation is used to analyze the
dynamics of this system, which leads to a second-order
underactuated model given by

(2)

Comparing with the model presented in [23], here we

ignore the mass of rotor and take the viscous resistance of
ground into account, which is inevitable in the reality. In
addition to the unknown slope angle , other model
uncertainties come from the body parameters m1 and I 1 .
For instance, if the MWIP model is used to describe a Segway
vehicle, then the body comprises of the control shaft and the
passenger. Owing to the different weights of persons, body
parameters m1 and I1 are apparently uncertain before the
controller design. The aim of this study is to design a robust
controller, which can make the vehicle move at a desired
velocity without steady error as well as guarantee the system
stability even there are model uncertainties and disturbances.
It should be pointed out that only straight velocity control
problem is investigated here. Hence three-dimensional model
of the vehicle as given in [1] is not required.
B. Analysis of equilibriums
By
choosing
the

x = [x1

x2

state

variable

as

x3 ] = 1 &w &1 , the state model of an

T

MWIP system can be represented by

x&1 = x3
M 11 x&2 + (M 12 cos( x1 + )) x&3 =

(3a)

~ (3b)

Dw x2 + D1 ( x3 x2 ) + M 12 x3 2 sin ( x1 + ) M
(M 12 cos(x1 + ))x&2 + M 22 x&3 =
D1 ( x3 x2 ) + G1 sin x1

(3c)

The vector form of the state model is given by

M (x) x& = F(x) + u

where

0
0
1

M 11
M 12 cos( x1 + ) ,
M ( x ) = 0
0 M 12 cos( x1 + )

M 22

2984

(4)

x3

~,
2

F(x) = Dw x2 + D1 ( x3 x2 ) + M 12 x3 sin ( x1 + ) M

D1 ( x3 x2 ) + G1 sin x1

u = [0

] T .

x2
Suppose x = x1
of system (4), satisfying

x3 = 0,

x3

is the desired equilibrium

G1 sin x1 = Dw x2 + M .

M 1 = M 11 + M 12 cos( x1 + ),

M 2 = M 22 + M 12 cos( x1 + )

x&1 = x3

M x& 2 + b(x ) = M 2
M x& + a (x ) = M
3
1

In the case of a velocity control problem, the desired

where

is then determined by (5).

inclination angle of the body,
Whereas, different body parameters and unknown slope

+ (M 11 + M 12 cos( x1 + ))D1 x3

+ (M 12 x3 ) cos(x1 + )sin ( x1 + )
2

~
M 11G1 sin x1 M 12 cos( x1 + ) M ,
b(x ) = [M 22 (D1 + Dw ) + M 12 D1 cos( x1 + )]x2
(M 22 + M 12 cos(x1 + ))D1 x3
M 12 M 22 x3 sin ( x1 + )
2

~
+ M 12G1 sin x1 cos( x1 + ) + M 22 M .

III. SLIDING MODE CONTROLLER DESIGN

controller design scheme is proposed in this section.

Note that variable M is positive because it is the

determinant of the positive definite matrix of the kinetic
energy [22].
Define

A. Controller Design
We start by introducing a new state vector

z = [z1

z2

z3

z4 ]

f ( x) =

*
= 1 &1 w &w t &w &w

].

* T

(6)

The sliding surface is given by

s = 1 z1 + z 2 + 2 z3 + 3 z 4

where coefficients

1 , 2 , 3

f f F .

(7)

(8)

where is a positive constant.

From (3b) and (3c), it follows that

~
2
Dw x2 + G1 sin x1 + M 12 x3 sin ( x1 + ) M

(13)

Further, we assume that

are constants to be

ss& s

a(x) + 3b(x)
a (x) + 3b(x)
, f ( x) =
(12)
M
M

and the estimation error caused by modeling uncertainty and

disturbances is assumed to be bounded by some known
function F(x),

3 M 2 M 1

determined.
The state trajectories can be driven to the sliding surface by
choosing the control law such that

M 1 x& 2 + M 2 x&3 =

M = M 11M 22 (M 12 cos( x1 + )) ,

a(x ) = [M 12 (D1 + Dw ) cos( x1 + ) + M 11 D1 ]x2

angle result in different inclination angle x1 , which makes it

impossible to compute the exact equilibrium before the
controller design. Normally, it is difficult to design a
controller if we do not know the exact equilibrium in advance.
Especially in the case of using a linearized model, we even
cannot obtain such model because it is generated by
linearizing the nonlinear system around the equilibrium.
In the rest of this paper, denotes that the terms are
evaluated based on parameters of the nominal system moving
on a flat ground without any uncertainties and disturbances.

To ensure an MWIP system can move at a desired velocity

(11)

x1 ,

&w , a special sliding surface and the corresponding SMC

(10)

Eq. (9) represents the internal dynamic of the MWIP model,

which is independent on the control law . From (9), we can
rewrite (3) as

(5)

velocity &w = x2 is always given in advance. The desired

= (1 + )

3 M 2 M 1
M

, < 1 .(14)

Theorem 1. The achievement of a sliding motion on the

surface (6) can be guaranteed by selecting the control law

1M
1
a (x) + 3 b(x)
z2
3 M 2 M 1
3 M 2 M 1
~
M
2 M
(15)

z4 +
k sgn( s )
M M
M M

=
(9)

where

2985

~
where a ( x) and b( x) are given by (11), and k , ~
r satisfy

F + ~
r + ~
~
k
, r = f 1 z 2 2 z 4 . (16)
1

Proof.
Using the equivalent control method, the equivalent
control law can be easily obtained as

1
a (x) + 3 b(x)

3 M 2 M 1
1M
2 M

z2
z4 .
(17)
3 M 2 M 1
3 M 2 M 1
Choose the sliding control law as = + d , the rest
work is to find appropriate switching control law d .

B. Surface Stability Analysis

The controller (15) ensures that all system trajectories will
reach the surface and remain there in the presence of model
uncertainties and disturbances. The next step is to find
appropriate coefficients 1 , 2 , 3 that guarantee the
trajectories are asymptotic stable during the sliding phase. On
the sliding surface, the closed-loop system dynamics is
represented by the combination of the sliding surface in the
form s = 0 and the internal dynamics given by (9).
Choosing a new state vector

q = [q1

1 2

(20b)
(20c)

(q) = [1 (q) 2 (q) 3 (q)] ,

T

1 (q) = q2 ,
2 (q) =

3 M 2 M 1

[1M 1q2 + ( 2 M 1 3 Dw )q3 +

(22)

~
3 M 12 q2 sin (q1 + ) + 3G1 sin q1 Dw 3&w 3 M ,

3 (q) =

3 M 2 M 1

[ 1M 2 q2 ( 2 M 2 Dw )q3

~
2

M 12 q2 sin (q1 + ) G1 sin q1 + Dw&w + M .

Obviously point

*
1

q = x
*

0 0 is one of the
*

equilibriums of the nonlinear system (21), where x1 is the

real desired inclination body angle determined by (5). Our
aim is to choose suitable coefficients 1 , 2 , 3 that

M M1
+ 3 2
d
M

guarantee the asymptotical stability of this equilibrium.

Choose

d =
~

where

2 4

z4 ] ,

Eqs. (20) can be further transformed into the following

vector form:
q& = (q)
(21)

3 M 2 M 1

a (x) + 3b(x) [a(x) + 3b(x)]

3 M 2 M 1

M M M

1
3 2
1 (1 z 2 + 2 z4 )
M 3 M 2 M 1

M M1
+ 3 2
d
M

= f f + f z z

z2

*
M 1q&3 + M 2 q&2 = Dw q3 + &w + G1 sin q1
~
2
+ M 12 q2 sin(q1 + ) M ,
1q2 + q&2 + 2 q3 + 3q&3 = 0 .

s& = 1 z&1 + z&2 + 2 z&3 + 3 z&4

1
=
M

the closed-loop system dynamics on sliding surface (7) is

represented by
q&1 = q2 ,
(20a)

From (6), (11) and (17), the derivative of s is given by

a(x) M 1 ( + d )
= 1 z&1 +

+ 2 z4
M
M

b(x) M 2 ( + d )
+ 3
+

M
M

1
= 1 z&1 + 2 z 4 + [ a(x) 3b(x) + 3 M 2 M 1]
M
M M1
+ 3 2
d
M

q3 ] = [z1

q2

~
M
k sgn( s )
3 M 2 M 1

(18)

where k = k ( x ) is a state-dependent function and sgn() is

the signum function. It follows that

[(

~
ss& = s f f + ~
r + ( + 1) k sgn( s )
~
s F + ~
r + (1 )k .

Theorem 2. For the MWIP system (4), design sliding

surface (7) and control law (15). Assume inequalities

M 1 (q * ), M 2 (q * ) > 0 and cos( x1 ) > 0

coefficients 1 , 2 , 3 satisfy

if

1M 1 (q ) + Dw > 2 M 2 (q ),

(19)

3 M 2 (q ) M 1 (q ) < 0,
1 , 2 , 3 > 0, 13 > 2

If k has the bound defined by (16), then the sliding

condition (8) is satisfied.

hold,

(23)

then the closed-loop system is locally asymptotic stable

2986

around equilibrium

parameters. An LQR controller designed by linearizing

nominal system was found cannot stabilize the actual system.

q* .

Proof.

TABLE I
EXACTLY KNOWN PHYSICAL PARAMETERS OF THE MWIP SYSTEM

We linearize system (22) about the equilibrium q to

establish linear stability criteria that guarantee local
exponential stability of the nonlinear system. The linearized
system is described by

q& = B (q q* )

(24)

where

B=

1 =

2
q1

q q

0
= 1
4

, 2 =
q

3
4 =
q1

2
q2

1
2
5

0
3 ,
6

, 3 =
q

3
, 5 =
q2
q

2
q3

3
, 6 =
q3
q

(25)

1 6 3 4 > 0,

(27)

It follows from (22) and (26) that

1 2 + 3 4
where

* 2

*
1

2 )M 1 (q* ) + 3 Dw ,

is given by

* =

I w = 0.6

r = 0.254

Length between
the wheel axle
and the center of
gravity of the
body (m)

viscous resistance in
the driving system
(N.s/m)

viscous
resistance of the
ground (N.s/m)

l = 0.267

D1 = 0.1

Dw = 3.3

Mass of the
body (Kg)

Moment of inertia of
the body (Kg.m2)

Actual system

m1 = 310.6

I1 = 65.0

Nominal system

m 1 = 137.6

I1 = 35.0

The function F in (13) and the parameter in (14) were

selected as 0.1 f and 0.1, respectively. Assume the

( )
(
= ( ) G cos x [(

m w = 29.0

(26)

2 + 6 = 1 * M 1 (q* ) + 2 * M 2 (q* ) * Dw ,
3 5 2 6 = * 1Dw 3 M 2 (q* ) M 1 (q* ) ,

Radius of the
wheel (m)

TABLE II
UNCERTAIN BODY PARAMETERS OF THE MWIP SYSTEM

( 2 + 6 )(35 2 6 ) + 1 2 + 3 4 > 0
1 6 3 4 = 2 *G1 cos x1 ,

Moment of inertia of
the wheel (Kg.m2)

According to the Hurwitz stability criteria, the equilibrium

of linearized system (24) is asymptotic stable if we have

2 + 6 < 0,

Mass of the
wheel (Kg)

selected as

1 = 1.5, 2 = 0.01

different cases of ground slope

1
M 1 (q ) 3 M 2 (q )

= 5

Apparently inequalities (27) hold if coefficients

1 , 2 , 3 are chosen to satisfy (23). This completes the
proof.

Note that the problem of uncertain equilibriums is solved

by using a special sliding surface (7). The third item of (7)
plays an important role in avoiding uncertain equilibriums. In
fact, state z3 is the integration of state z 4 , which can
eliminate the static tracking error like a integral part in a
common-used PID controller.
IV. SIMULATION STUDY
Consider an MWIP system whose exactly known system
parameters are list in Table I. To test the robustness of the
proposed controllers, it is assumed that the actual physical
parameters of the body are different from that of the nominal
system. Table II shows the actual and nominal body

desired moving velocity is x 2 = &w = 7 rad / s . In order to

avoid chattering associated with the sliding mode control law,
we has approximated the discontinuous sign function (sgn(s))
with continuous saturation function of small boundary layers.
According to Theorem 2, coefficients 1 , 2 , 3 were
and

3 = 0.05 .

= 0o

Three

= 5o

and

were considered in the simulation.

The final simulation results using the proposed SMC
controller are depicted in Fig. 2, 3 and 4. In all the three cases,
the MWIP system is asymptotical stable with very small
steady tracking error, which substantiates the efficiency of
the presented approach.
V. CONCLUSION
The robust velocity tracking control problem of mobile
wheeled inverted pendulum models is investigated in this
study. Sliding mode control method is chosen due to its
robustness to nonlinear systems. Normally, it is hard to
design an SMC controller for general underactuated
mechanical systems. In our special case, we proposed a new
SMC controller for the mobile wheeled inverted pendulum
considering both parameter uncertainties and external
disturbance. The proposed SMC controller can eliminate the
steady tracking error by using a special sliding surface. The
efficiency of the proposed controller is confirmed by

2987

[2]

numerical simulation.

[3]
[4]

[5]
[6]
[7]
[8]

Fig. 2.

System trajectories by using the SMC controller when

= 0o . (Dot line denotes the desired velocity)

[9]
[10]

[11]
[12]
[13]
[14]

[15]
Fig. 3.

System trajectories by using the SMC controller when

= 5 o . (Dot line denotes the desired velocity)

[16]
[17]
[18]
[19]
[20]
[21]

Fig. 4.

System trajectories by using the SMC controller when

= 5 o . (Dot line denotes the desired velocity)

[22]
[23]

REFERENCES
[1]

K. Pathak, J. Franch, S.K. Agrawal, Velocity and position control of a

wheeled inverted pendulum by partial feedback linearization, IEEE
Trans. Robotics, vol. 21, no. 3, pp. 505- 513, 2005.

2988

A. Salerno, J. Angeles, The control of semi-autonomous two-wheeled

robots undergoing large payload-variations, in Proc. IEEE Int. Conf.
on Robotics and Automation, New Orleans, LA, 2004, pp. 1740-1745.
Y.-S. Ha and S. Yuta, Trajectory tracking control for navigation of the
inverse pendulum type self-contained mobile robot, Robot. Autonom.
Syst., vol. 17, pp. 6580, 1996.
M. Baloh and M. Parent, Modeling and model verification of an
intelligent self-balancing two-wheeled vehicle for an autonomous
urban transportation system, in Proceeding. of Conf. Comp.
Intelligence, Robot. Autonom. Syst., 2003, Singapore.
F. Grasser, A. DArrigo, S. Colombi, and A. Rufer, Joe: A mobile,
inverted pendulum, IEEE Trans. Ind. Electron., vol. 49, no. 1, pp.
107114, 2002.
Available: http://www.geology.smu.edu/~dpa-www/robo/nbot/
Available: http://www.teamhassenplug.org/robots/legway/
H. Tirmant, , M. Baloh, L. Vermeiren, T. M. Guerra and M. Parent, B2,
an alternative two wheeled vehicle for an automated urban
transportation system, in Proceeding of IEEE Intelligent Vehicle
Symposium, Paris, France, 2002, pp. 594603.
Available: http://www.segway.com
M. Karkoub, M. Parent, Modelling and non-linear feedback
stabilization of a two-wheel vehicle, Proceedings of the Institution of
Mechanical Engineers, Part I: Journal of Systems and Control
Engineering, vol. 218, no. 8, pp. 675-686, 2004.
M. W. Spong, The swing up control problem for the acrobat, IEEE
Control Systems Mag., vol. 15, no. 1, pp. 4955, 1995.
I. Fantoni, R. Lozano, and M. W. Spong, Energy based control of the
pendubot, IEEE Trans. Autom. Control, vol. 45, no. 4, pp. 725729
2000.
Y.-L. Gu, Direct adaptive control scheme for underactuated dynamic
systems, in Proc of the IEEE Conference on Decision and Control,
San Antonio, Texas, 1993, pp. 16251627.
A. M. Bloch, N. E. Leonard and J. E. Marsden, Stabilization of the
pendulum on a rotor arm by the method of controlled Lagrangians, in
Proc. of the IEEE International Conference on Robotics and
Automation, Piscataway, New Jersey, 1999, pp. 500505.
M. Karkoub, Modelling and robust -synthesis control of an
intelligent self balancing two-wheel vehicle, Proceedings of the
Institution of Mechanical Engineers, Part K: Journal of Multi-body
Dynamics, vol. 220, no. 4, pp. 293-302, 2006.
W. Wang, J. Yi, D. Zhao and D. Liu, Design of a stable sliding-mode
controller for a class of second-order underactuated systems, IEE Proc.
Control Theory Appl., vol. 151, no. 6, 683-690 (2004).
B.L. Ma, Comment on design of a stable sliding-mode controller for a
class of second-order underactuated systems, IET Control Theory
Appl., vol. 1, no. 4, 1186-1187, 2007.
H. Ashrafiuon, R. S. Erwin, Sliding control approach to underactuated
multibody systems, in Proc. American Control Conference, Boston,
MA, 2004, pp. 1283-1288.
M. Nikkhah, H. Ashrafiuon and F. Fahimi, Robust control of
underactuated biped using sliding modes, Robotica, vol. 25, no. 3,
367-374, 2006.
K. Lee, V. Coverstone-Carroll, Control algorithms for stabilizing
underactuated robots, J. Robotic Sys., vol. 15, no. 12, pp. 681-697,
1998.
C.-Y. Su, Y. Stepanenko, Sliding mode control of non-holonomic
systems: underactuated manipulator case, in Proc. of the International
Federation of Automatic Control (IFAC) on Nonlinear Control Systems
Design, Tahoe City, California, 1995, pp. 609613.
A. M. Formalskii, An inverted pendulum on a fixed and a moving
base, J. Applied Mathematics and Mechanics, vol. 70, pp. 56-64,
2006.
O. Matsumoto, S. Kajita and K. Tani, Estimation and control of the
attitude of a dynamic mobile robot using internal sensors, Advanced
Robotics, vol. 7, no. 2, pp. 159-178, 1993.