You are on page 1of 6

Analysis and Control of an Omnidirectional Mobile

Robot
Chao Ren1 and Shugen Ma2
I Department

of Robotics, Ritsumeikan University, gr0119vp@ed.ritsumei.ac.jp


of Robotics, Ritsumeikan University, shugen@se.ritsumei.ac.jp

2 Department

Abstract- In this paper, the control system design is presented

for an omnidirectional mobile robot based on MY wheels-II.

First, the MY wheel-II mechanism and prototype platform


are briefly introduced.

Then the detailed dynamic model is

derived, which shows the robot is a switched nonlinear system.


To deal with the difficulties and challenges in the controller
design for the switched nonlinear system, a simple but robust
control system design, based on active disturbance rejection
control framework, is proposed. The proposed control system
requires very little robot information.

The basic idea is to

estimate the unknown internal changing dynamics of the robot


and external disturbances by the extended state observer, and
then to actively compensate them by the control signal in real
time.

In addition, the stability properties are also analyzed

based on the previous analysis results. Simulation results show


the effectiveness and strong robustness of the proposed control
method.

Index Terms- Active disturbance rejection control, omnidi


rectional mobile robot, robust control, switched nonlinear sys
tem.

I. INTRODUCTION
In the past few years, omnidirectional mobile robots
(OMRs) have received growing attention in the field of
wheeled mobile robots because of their high maneuverability.
Compared with non-holonomic mobile robots, OMRs, which
are holonomic, are able to achieve translational and rational
motions independently and simultaneously. As a result, they
are extremely useful in tight environments, such as hospitals,
warehouses, residential homes, and sheltered workshops for
disabled people.
A variety of omnidirectional wheel mechanisms have been
proposed over the past few decades. Several mechanisms
based on the "universal wheel concept" have been designed,
which is an assembly providing a combination of constrained
and unconstrained motion when turning [1]. The initial
universal wheel design was used to accomplish omnidirec
tional motions without changing the direction of the wheels,
but this wheel type suffers from successive shocks when
the individual rollers make contact with ground. To realize
smoother contact with the ground, the "orthogonal-wheels"
concept and two major wheel assemblies, i.e., Longitudi
nal and Lateral orthogonal-wheel assemblies, were proposed
in [1]. Based on the same idea but with different number
and slicing of the spheres, MY wheel [2] and MY wheel
II [3] were proposed. The previous MY wheel may get stuck
by surface irregularities due to the specific wheel structure,
i.e., the gaps between each two adjacent crowns are rela
tively large. To overcome this problem, the MY wheel-II

was proposed with only 10 gap between two adjacent spher


ical crowns. In addition, compared with other omnidirec
tional wheel structures, such as Mecanum wheel [4], Swedish
wheel [5], this wheel mechanism is relatively simple and sta
ble, insensitive to dirt and fragments on the floor, and has a
higher payload carrying ability.
However, the MY wheel mechanism has two alternate con
tact point with the ground, i.e., inner wheel contact point and
outer wheel contact point. In practice, we found it difficult to
design the dynamic controller for the robot, because the robot
is a switched nonlinear system due to the wheel switching.
The control system design for the switched nonlinear system
is a great challenge. In addition, all of the previous dynamic
modeling and control designs were for the OMRs with con
tinuous dynamics, such as [6-8]. It should be pointed out that
although Swedish wheel also has switching contact points,
continuous dynamic models were directly employed in all of
the previous controller design studies by considering it as a
non-switch wheel. This is because, the space between the two
small wheel parts is usually small compared with the contact
radius, and the switching effects can be neglected. However,
for the same size robot, the discontinuity of OMRs on the
basis of MY wheels-II becomes much more pronounced, be
cause of the large space between the two small wheel parts.
Motivated by the challenges and difficulties in the con
trol design for switched nonlinear system, in this paper, we
propose a simple control system design for the robot based
on active disturbance rejection control (ADRC) framework.
ADRC is a novel and inherently robust control method while
requiring very little information of the plant dynamics [9].
ADRC has been successfully applied in various engineer
ing applications, e.g., motion control [10-13], robot con
trol [14,15], DC-DC power converter control [16].
In this paper, the MY wheel-II mechanism and a three
wheeled prototype platform are firstly introduced. The dy
namic properties of the robot are analyzed. Then ADRC is
employed in the control system design, in which the internal
dynamics (including the input-output cross coupling effects
and the changing dynamics due to the switching between sub
systems) and external disturbances can be approximately es
timated by extended state observer (ESO) and compensated
in the control input. However, the robot switching dynamics
is failed to be accurately estimated by ESO and the resulting
effects on the robot motion are then analyzed.
The remainder of this paper is organized as follows. In
Section II, the MY wheel-II mechanism and prototype plat
form are introduced. In Section III, the kinematic and dy
namic model, as well as the analysis, are presented. The
ADRC design for the robot is presented in Section IV. In Sec
tion V, simulation results, analysis and discussion of the pro-

(a)
Fig.

(b)

1. (a) End view of MY wheel-II. (b) the MY wheel-II assembly.

posed control design are presented. Finally, conclusions are


drawn in Section VI.
II. MY WHEEL-II MECHANISM AND ROBOT
PLATFORM
The end-view of the MY wheel-II mechanism and the as
sembly are shown in Fig. l. The wheel consists of two balls
of equal diameter on a common shaft and both balls are sliced
into four spherical crowns. Each spherical crown can rotate
freely around its axis. The two sets of spherical crowns are
mounted at 45 from each other to produce a combined cir
cular profile. During the rotation of main shaft, the two sets
of crowns make alternate contact with the ground to realize
continuous motion on the ground. The two contact points
with ground switch between the two sets of crowns whenever
the shaft turns 45 , and therefore eight switches occur dur
ing each turn (see Fig. lea. Attention should be paid to the
absolute encoder installed in each assembly, which is used to
detect the switches of the contact radius (see Fig. l(b. If
odometry is employed in the positioning of switch wheeled
OMRs, the absolute encoder is indispensable. The prototype
platform is shown in Fig. 2, with three MY wheel-II assem
blies arranged with a 120 interval angle underneath the steel
disk.
There are two main advantages of the MY wheel-II mecha
nism over the traditional wheels. One is that the MY wheel-II
mechanism has large payload capacity compared with the tra
ditional wheels, such as Swedish wheel and Ball wheel. This
can be easily obtained from the MY wheel-II mechanism.
The other advantage is that the MY wheel-II mechanism is
much more insensitive to the fragments on the floor, such the
wool. The small rollers in the Swedish wheel or Mecanum
wheel can be easily gotten stuck, because the small rollers
may be twined by fragments on the ground (i.g., wool). How
ever, the MY wheel-II, due to fact that the rotation shaft of the
spherical crown is located inside the wheel, can not be easily
gotten stuck. In addition, the MY wheel-II has a higher step
climbing capability (average of 20mm height with a 60mm
radius of the wheel). For a detailed description of the MY
wheel-II mechanism and the prototype platform, please refer
to [3].
However, one drawback of this wheel is that the contact
radius switches between the inner and outer wheel radius, re
sulting in a switched nonlinear robot system. The control
design for the switched systems is usually challenging and
complicated.
III. MODELING
There are two coordinate frames used in the modeling (Fig.
3): the world coordinate frame {W} fixed on the ground and

Fig.

1>2

\\\
....

2. Prototype platform

D,

.
. ...... ./
.

.
.
.
..
..
.
..
. . ...... ....
..
.
. ..
.. .
.. .
...
.
......
.. .
..
.... ...
.
.

.
..
..
.

Fig.

3. Coordinate frames of the omnidirectional mobile robot.

the moving coordinate frame {M} fixed on the robot gravity


center.
The coordinate transformation matrix from the moving co
ordinate frame to the world coordinate frame is as follows:
-sin8
cos8
o

A.

(1)

Kinematic Modeling

The inverse kinematic equation of the mobile robot in the


world coordinate can be expressed as follows [3]:
(2)

(3)
where
-cos( -8)
sin(8- )
cos8
W

=
=

WI
y

W2

W3

is

cos( +8)
-cos( -8)
sin8
the

motor

angular

rate,

is robot location and orientation an

is the angular
gle in the world frame, cD = (P I (P2 (P 3
velocity of the wheel shaft, n is the gear reduction ratio, R is

the wheel radius, Li is the contact radius of each assembly,


i=1,2,3,

Ll.

{ D'Doub ifIf
t n,

!!.

+ !:!:!!.
2 <
1C
nrr
8 + 2'

A..
'f/l

<

<
-

l!!.
me
8 + 2
IT
nn
::; 8 + 2'

n=0,I,2,...

(4)
and
D
t
ou are the inner and outer contact radius, re
and Din
spectively; i is the angular position of the wheel shaft.
B.

Dynamic Modeling

The detailed dynamic modeling was presented in our pre


vious work [17]. The robot dynamic model in the world co
ordinate frame can be obtained as [17]:

Mq +Cq=Bu

po + m

(5)

IV. ADRC DESIGN


The ADRC design block diagram for the mobile robot is
shown in Fig. 4, which consists of two parts: controller and
ESO.

A.

Controller Design
The robot dynamic model (5) can be rewritten as:

where wet) is a 3 x 1 vector representing the unknown factors,


such as dynamic variations, uncertainties, state-dependent
nonlinearities and external disturbances. The term Mo 1 Bo is
the approximation of M IB and is the only required informa
tion about the robot in the ADRC design. In addition, MoIBo

o
Ll+Lr2L3.
)
Po (
2

without rotation. The nonlinearity and discontinuity are in


troduced into the robot dynamics when the robot rotates. In
other words, when the robot performs only translational mo
tion, the kinematic and dynamical properties of the robot
are the same as the robot with non-switch wheels, such as
Mecanum wheel and Alternate wheel [19].

po + m

V; po(LI - L2)

does not need to be known exactly, as the resulting error can


be also rejected in ADRC as one of the perturbations.
Define
fU)= M-I(Cq + w(t))+(M IB _MOIBo)u

-me

PI

V; Pl (LI - L2)

[ fxU) fy(t) feU) r

where f( t) =
is regarded as the 'to
tal disturbances' of the three channels, including the input
output cross coupling effects and the changing dynamics of
the robot due to the switching.
And then (6) can be written as

q=fU) +MOlBou
Po

_
-

n2/0

R2' P 1

_
-

n2

R2

ktk,,
(b 0 + To) ' P2

r,

nkt

- RRa'

= UI U2 U3
and Ui is the applied motor voltage, m is the robot mass, Ra is
the armature resistance, kb is the motor back emf constant, k/
is the motor torque constant, Iv is the robot moment of inertia,
Iw is the wheel moment of inertia, 10 is the combined moment
of inertia of the motor, gear train and wheel referred to the
motor shaft, bo is the combined viscous friction coefficient of
the motor, gear and wheel shaft.
Since the contact radius L; switches between the inner and
outer wheel contact radius, i.e., between D;n and Dout (see
(4)), it can be seen from (5) that the robot is a switched non
linear system. In addition, as each wheel has two contact
modes, a three wheeled OMR based on MY wheels-II has
eight contact modes in total. Thereby, the robot has eight
smooth nonlinear subsystems in total [17]. However, not all
of the eight contact modes always appear in the robot motion.
It depends on the robot initial orientation, wheel initial posi
tion and robot trajectory, which can be derived from the robot
kinematic equation (2). Moreover, the switching signal of
the switched robot system is the contact radius L;, i= 1,2,3.
The contact radius depends on the the wheel angular posi
tion, while the latter depends on the robot trajectory (i.e., the
robot states). Therefore, the mobile robot is an autonomous
switching system [18].
In addition, it can be seen from (5) that the mobile robot
will be a smooth linear dynamical system if the robot moves

(7)

Defining C(t) is the estimation of f( t) , we apply the control


law:

(8)
Define the estimation error of the 'total disturbances' as:

fe(t) =f(t) - C(t)

(9)

[ faCt) fey(t) fee(t) r

where fe(t) =
With the assumption that f(t) can be well estimated, i.e.,
fe(t) 0, then the nonlinear piecewise-smooth MIMO robot
dynamics can be reduced to three decoupled SISO double in
tegrators in an approximate manner:

(10)
For the decoupled double integral systems (10), simple PD
control is applied to the three channels:

[ :r

[ :

The gai

(11) can be selected

Kp =

, Kd =

(11)

e
2we

2we
w
where We is the controller bandwidth [20]. The selection of
Kp and Kd is to make the characteristic polynomial of each
channel Hurwitz, i.e., s2 + 2wc + w is Hurwitz.

q=[XYB]T

Fig.

Stability Analysis

Although ADRC has demonstrated validity and advantages


in many applications, the stability of ADRC is studied in only
few papers in recent years. The up-to-date analysis results
were shown in [21] for the nonlinear MIMO system, which
can be formulated as follows:

l. Convergence of ESO. Assuming g(t) (see (13 is


bounded, the estimation error of the ESO is bounded and
its upper bound monotonously decreases with the observer
bandwidth.

4. ADRC for the mobile robot.

Combining (8) and (11), the final control law is:

We,

For the controller (12), only one parameter, i.e.,


is to
be tuned. The high controller bandwidth will improve the
closed-loop system response. However, the system will be
more sensitive to the noises, leading to oscillations or even
instability. Therefore, a trade off should be made between the
closed-loop performance and the stability margin and noise
sensitivity.
B.

C.

ESO Design

The estimation of f(t) is indispensable for the implemen


tation of control law (12), which will be accomplished by
ESO. ESO is a kind of disturbance estimator, which estimates
the combined unknown internal dynamics and external distur
bances in addition to the states in the real time by adding an
extended state [9].

2. Convergence of ADRC. Assuming get) is bounded,


the closed-loop tracking error of ADRC is bounded and its
bound monotonously decreases with the observer and con
troller bandwidth.
Theoretically speaking, for the switch wheeled OMRs, get)
is radially unbounded in the states at the discrete switching
instants. However, as discussed in [21], g(t) is physically
bounded in practice since it represents the jerk of the robot.
Therefore, the stability analysis results in [21] can be applied
to our robot in practice.
V.

SIMULATION AND DISCUSSIONS

Define Xl =
y e
X2 =
Y e
X3 = f(t).
X3 is the augmented state vector, the robot dynamic model
(7) can be written in the state space form:

In this section, the simulation of the control system is im


plemented in Matlab/Simulink. The parameter values used in
the simulation are as follows: m = 33 kg, Iv = l.35 kg m2,
R = 0.06 m, Din = 0.147 m, DOlll = 0.236 m, 10 = 9.7 X
10-6 kg m2, k, = 0.0292N miA, kb = 328 rpm/V, n = 26,
bo = 6 x 10-5 Nms / rad, Ra = 0.61 Q.

Xl =X2
.
X2 =X3 +M-LB
0 OU
X3 = get)

(13)

approximation of M-lB. The calculation process of the term


M0 1 Bo is as follows:

where g(t) represents the rate of change of the acceleration,


i.e., the jerk, which is physically bounded in practice.
Define Zi, i = 1,2,3, as the estimation of state Xi. Then a
linear ESO is designed for (13) as follows [20]:

l. Some parameter values in the matrix M are chosen as


40% less than the real ones to calculate Mo, including m, Iv
and 10.
2. The varying contact radius Ll, L2 and L3 are chosen as
the average one in the matrix B to obtain Bo:

[x

r,

[i

el = Zl -XI
ZI=z2-/30 1 el
zFz3-/302el +M01Bou
ZF -/303el

r,

D +D

Ll =L2 =L3 =.
(14)

The observer gains can be chosen as:

POI =

P02 =

[
[

P03 =

Wo

3w

As mentioned before, the only information about the robot


required in the ADRC design is the term MoLBO, i.e., the

where
is the observer bandwidth [20] and is the only pa
rameter to be tuned in ESO. In addition, the high observer
bandwidth will improve the system response and estimation
performance of the ESO but subject to noise sensitivity. In
practice, therefore, a trade off also exists in the parameter
tuning of ESO.

We

wo)

The tuning method of the controller and observer param


eters (i.e.,
and
was discussed in details in [22]. A
cornmon rule of thumb is to choose:
(3
In
the simulation, the controller and observer parameters are
set as:
= 3,
= 15. The saturation voltage of the
motor is 12 V. The robot initial position and orientation

We

Wo

Wo

5)we.

is set as
Om Om Orad ion voltage of the motor is
12 V. The robot initial position and orientation is set as
Om Om Orad

The robot is commanded to track a circle of 1 m radius at


an angular rate of Is rad/s. In the first 10 s, the robot per
forms translational motion without rotation. After 10 s, the
robot moves with rotation at a desired rotational angular ve
locity of 3 rad/s. After 20 s, to investigate the robustness
of ADRC against the robot parameter variations and external
disturbances, the robot viscous friction coefficient bo is made
to change as times of the original one. The robot inertial
parameters, the robot mass m and moment of inertia Iv, are
made to change as 10 times of the original ones. The sim
ulation results are shown in Fig.
Fig. 10. Analysis and
discussion of the simulation results are as follows:

5-

1-

Rererence- Robol
- - Observed

:E;;--
:P S;-d 1'r><'.:-
;
<=
= ----""-; -------+.:""---Jl
:; t j : t
0.8

-X

10

15

10

20

15

25

20

30

25

t(8)

Fig.

.q;

5. Robot tracking response.

1,:D f ==-= ii
, ii
l I=
I t .
;0

-";:5

30

..L,0=--

;0

--

--

,0=--

g
:3

---

I(S)

2.8

....'

1.6

N'

3.5 5

:3

3.45

4-<'

7.4

t(8)

25

---

"i:_

30

t(8)

f(t) and its estimation r(t) .

0
,0
0

'
5

10

Fig.

15

20

25

30

.
10

10

15

20

25

30

15

20

25

30

1(8)

Fig.

25

....A.

30

c-n<'

20

25

30

7. Applied control input: motor voltage

It can be seen form Fig. 5 and Fig. 10 that the tracking


performance of the proposed ADRC design is excellent with
small steady-state tracking error. In addition, it is shown in
our simulations that the increase of the bandwidth of con
troller or observer will improve the transient and steady-state
tracking performance, but it may suffer from the noises in the
practice.
Moreover, as shown in Fig. 6, for the robot moving with
out rotation, the robot dynamics is smooth (see the partial en
larged view A). However, the sharp changes in the robot dy
namics at the discrete switching instants are introduced when
the robot moves with rotation (see the partial enlarged view

8.4

t(8)

B.6

-0.21,'c;71 ----,;;
,8-;-:
7-.;;
8 ---- ,-;C;7-;;
-c 7;c;.2 ----:-17C:-CAc
.6----,-: C:

B.8

t(8)

.1

q.

0.2
0.15
0.1

''-

0.2
B.2

8. Reference velocity itd, robot velocity it and its estimation

\;-----+----;-;;--+._---c--!c_---L

0.25 ___"""CT"---------- --
0.2

0.15
0.1

;;----7---7;;---;-;--.;;;---;'c--- .

0.2
0.15
0.1

\;-3---"o---:-::---:-:,5c-::20:---::25=--- 0

Fig.

'

...L2=5
-

'

--

6. The 'total disturbances'

20

15

<:'IWIt'i:N>
o: ====-;:===;1_,' '___
+.c-J_",'_ _'
o
10
15
t(8)

-0.08 8

2.4

Fig.

10

0.06

30

1.2 '-:----:-::--::---:-:--:------:-::-::---:---:-!
-:-:
12
12.2 12.4 12.6

7.8

7.6

:-:: 20

"1

3.65

".-

-0.04

L-...L2
5
=:..:.:.JI30
:::

20

---',5

-0.40

.,.,

--

:-::
20

15

3.75

---:;;-',5

10

---

---

--

---

---

_.

0.16c------

10

20

'\
o '----i

0
,
'0
A

__ _ ,
.0.3

nn----
-
----,0
, 5-----.:20
25--
305
0

30

10 ;---
------':;;-:O ----:-:
15C------.:2::-----:-::
25-0
30

1\

OA /

t(8)

9. The contact radius of the three MY wheel-II assemblies.

B). Moreover, it is also shown in Fig. 6 that the exact robot


dynamics can not be well estimated by ESO when the robot
rotates. In addition, it can be seen from the switching signal (Fig. 9) that the switching between subsystems is very
fast. As a result, the ESO is not able to well estimate the 'to
tal disturbances' especially when the robot dynamics moves
with fast switching between subsystems. One resulting effect on the robot motion is that the robot velocity has slight
vibrations, as shown in Fig. 8.
In addition, as shown in Fig. 5 and Fig. 10, the performance of the controller maintains the same level after 20 s,
which demonstrates the strong robustness of the control system against large internal and external disturbances.
Finally, it is worth noting that the exact switching time
is not required in the proposed ADRC design. Therefore,
the absolute encoder is not required, reducing the cost of the
control system design. In addition, the feedback signal of the
proposed ADRC design requires only robot position informa
tion, and the velocity information is obtained from the ESO.
Thereby, the system design cost is further reduced since the
sensors measuring the velocity information are unnecessary.
VI. CONCLUSIONS
In this paper, the ADRC system design has been pre
sented for a three wheeled OMR on the basis of MY wheels
II, whose dynamics is switched nonlinear. Firstly, the MY
wheel-II mechanism and the prototype platform were briefly

[7] K. Watanabe,

t=10S

0.5

E
;;:

J.

/,

...,

-0.5

- . - Robot Trajectory

... . -

:;/

-1

-1

-0.5

and

'/

[8] Y. Liu, 1. 1. Zhu, R. L. W. II, and 1. Wu, "Omni-directional


mobile robot controller based on trajectory linearization, "

Robotics and Autonomous Systems, vol. 56, no. 5, pp. 461 -

479, 2008.
[9] 1. Han, "From pid to active disturbance rejection control, " In

dustrial Electronics, iEEE Transactions on, vol. 56, no. 3, pp.


900 -906, march 2009.
[10] S. Li and Z. Liu, "Adaptive speed control for permanent

-._.-'

-1.5

Fig.

Initial position

1. Tang,

tonomous platform for mobile service robots, " Journal of In


telligent and Robotic Systems, vol. 22, pp. 315-330, 1998.

Y. Shiraishi, S. Tzafestas,

T. Fukuda, "Feedback control of an omnidirectional au

'

x(m)

0.5

1.5

10. The reference trajectory and robot trajectory in xy-plane.

introduced, including the features of the MY wheel-II. The


dynamic properties of the robot were analyzed briefly. Then
ADRC was applied to the mobile robot, wherein the switch
ing nonlinear dynamics and external disturbances can be ap
proximately estimated by ESO and actively compensated in
the control signal. However, the fact that the exact robot
switching dynamics can not be well estimated by ESO leads
to slight robot vibrations. It is shown in the stability analysis
that the tracking error of the closed-loop system is bounded
in practice. Simulation results have demonstrated the effec
tiveness and robustness of the proposed design. In addition,
the main advantages of the proposed ADRC design are: i)
simple but with strong robustness, while requiring very little
plant information; ii) only two parameters to be tuned and the
parameter turning process is explicit and easy; iii) lower cost,
because the sensors measuring robot velocity and the exact
switching time are not necessary.
Currently, we are developing the robot navigation system,
and therefore experimental studies of the proposed control
method will be carried out in the near future.

magnet synchronous motor system with variations of load in


ertia, " industrial Electronics, iEEE Transactions on, vol. 56,
no. 8, pp. 3050-3059, 2009.
[11] Q. Zheng, L. Q. Gao, and Z. Gao, "On validation of extended
state observer through analysis and experimentation, " Journal

of Dynamic Systems, Measurement, and Control, vol. 134, p.


024505, 2012.
[12] S. Zhao and Z. Gao, "An active disturbance rejection based ap
proach to vibration suppression in two-inertia systems, " Asian
Journal of Control, 2012.
[13] H. Liu and S. Li, "Speed control for pmsm servo system using
predictive functional control and extended state observer, " In
dustrial Electronics, iEEE Transactions on, vol. 59, no. 2, pp.
1171-1183, 2012.
[14] Y. Su, B. Duan, C. Zheng, Y. Zhang, G. Chen, and 1. Mi,
" Disturbance-rejection high-precision motion control of a
stewart platform, " Control Systems Technology, IEEE Trans

actions on, vol. 12, no. 3, pp. 364 - 374, may 2004.
[15] S. Talole, 1. Kolhe, and S. Phadke, "Extended-state-observer
based control of flexible-joint system with experimental vali
dation, " Industrial Electronics, IEEE Transactions on, vol. 57,
no. 4, pp. 1411 -1419, april 2010.
[16] B. Sun and Z. Gao, "A dsp-based active disturbance rejection
control design for a l-kw h-bridge dc-dc power converter, " In
dustrial Electronics, iEEE Transactions on, vol. 52, no. 5, pp.
1271 - 1277, oct. 2005.

REFERENCES
[1] F. Pin and S. Killough, "A new family of omnidirectional and
holonomic wheeled platforms for mobile robots, " IEEE Trans
actions on Robotics and Automation, vol. 10, no. 4, pp. 480
-489, Aug 1994.
[2] C. Ye and S. Ma, " Development of an omnidirectional mobile
platform, " in Proceeding of IEEE International Conference on

Mechatronics andAutomation, Aug. 2009, pp. 1111 -1115.


[3] S. Ma, C. Ren, and C. Ye, "An omnidirectional mobile robot:

Concept and analysis, " in 20i2 IEEE International Conference


on Robotics and Biomimetics (ROBlO), Dec. 2012, pp. 920925.

[4] B. 1I0u, "Wheels for a course stable self-propelling vehicle


movable in any desired direction on the ground or some other
base, " USA Patent, 1975.
[5] G. Indiveri, "Swedish wheeled omnidirectional mobile robots:
Kinematics analysis and control, " Robotics, iEEE Transac
tions on, vol. 25, no. 1, pp. 164 -171, feb. 2009.
[6] I. Williams, R.L., B. Carter, P. Gallina, and G. Rosati, " Dy
namic model with slip for wheeled omnidirectional robots, "
Robotics and Automation, IEEE Transactions on, vol. 18,
no. 3, pp. 285 -293, jun 2002.

[17] c. Ren and S. Ma, " Dynamic modeling and analysis of an om


nidirectional mobile robot, " in Intelligent Robots and Systems

(iROS), 2013 IEEE/RSJ International Conference on, 2013,

accepted.
[18] M. Branicky, V. Borkar, and S. Mitter, "A unified framework
for hybrid control: model and optimal control theory, " Auto
matic Control, IEEE Transactions on, vol. 43, no. 1, pp. 3145, 1998.
[19] K.-S. Byun and I.-B. Song, " Design and construction of con
tinuous alternate wheels for an omnidirectional mobile robot, "
Journal of Robotic Systems, vol. 20, no. 9, pp. 569-579, 2003.
[20] Z. Gao, "Active disturbance rejection control: a paradigm shift
in feedback control system design, " inAmerican Control Con

ference, 2006, june 2006, p. 7 pp.

[21] Q. Zheng, Z. Chen, and Z. Gao, "A practical approach to dis


turbance decoupling control, " Control Engineering Practice,
vol. 17, no. 9, pp. 1016 - 1025, 2009.
[22] Z.Gao, "Scaling and bandwidth-parameterization based con
troller tuning, " in Proceedings of the American Control Con

ference, vol. 6, 2006, pp. 4989-4996.

You might also like