2008 IEEE International Conference on

Robotics and Automation

Pasadena, CA, USA, May 19-23, 2008

Navigation of an Omni-directional Mobile Robot with Active Caster

Wheels
Eui-jung Jung, Ho Yul Lee, Jae Hoon LEE, Byung-Ju Yi, Member, IEEE,
Whee Kuk Kim, Member, IEEE, and Shin’ichi Yuta, Member, IEEE

Abstract—This work deals with navigation of an wheel. Recently, researches on omni-directional mobile
omni-directional mobile robot with active caster wheels. robots with active caster wheels have drawn much attention.
Initially, the posture of the omni-directional mobile robot is Moore and Flann[8] and Berkermeier and Ma[9]
calculated by using the odometry information. Next, the position implemented a six-wheeled omni-directional mobile robot
accuracy of the mobile robot is measured through comparison and ODIS (Omni-Directional Inspection System) with the
of the odometry information and the external sensor
so-called “smart wheel” active caster wheel module,
measurement. Finally, for successful navigation of the mobile
robot, a motion planning algorithm that employs kinematic
respectively. Yi and Kim [13] initially developed the
redundancy resolution method is proposed. Through kinematic model of the omni-directional mobile robot with
experiments for multiple obstacles and multiple moving active caster wheels. Wada, et al.[10] developed the mobile
obstacles, the feasibility of the proposed navigation algorithm robot with two caster wheels and one rotational actuator.
was verified. Ushimi, et al.[11] developed an omni-directional vehicle with
two-wheeled casters. Lee, et al.[12] implemented and
I. INTRODUCTION proposed a motion generation algorithm for the caster-type

T HE types of wheels used in most of mobile robots can be

classified as four different types; conventional wheel,
centered orientable wheel, off-centered caster wheel, and
omni-directional mobile robot. Park, et al. [16] addressed the
optimal design for this type of mobile robots.
There are many researches on mobile robot navigation.
Swedish wheel[1-3]. For the mobile robot to have an However, navigation of the omni-directional mobile robots
omni-directional characteristic on the plane, only wheels with was rarely studied. Zavlangas, et al.[17] implemented this
three degrees of freedom must be employed in mobile robots. issue to a Swedish wheel type mobile robot, but it is not
Either the caster wheel or Swedish wheel can be modeled completely reliable because of accumulated odometry errors.
If odometry information is not reliable, the mobile should
kinematically as three degree of freedom serial chain. Chen,
depend on external sensor for navigation.
et al.[4] developed an omni-directional mobile robot using the
In light of these facts, this paper initially reviews a
crawler with free roller. Tadakuma, et al.[5] implemented an
kinematic model of the mobile robots with three active caster
omni-directional mobile robot including omni-disk using the
wheels. Then, we propose an algorithm to obtain reliable
passive wheel. Ziaie-Rad, et al. [6] developed Perisia
odometry information geometrically by multiple encoder
omni-directional soccer player robot with 6 omni-wheels.
information, and we also show the effectiveness of the
Salih, et al.[7] developed an omni-directional mobile robot
proposed algorithm through experimentation. Finally, we
using four custom-made mecanum wheels.
implement navigation algorithm to an omni-directional
However, it is known that either Swedish wheel or most of
mobile, which employs kinematic redundancy resolution
other type of “omni-directional wheels” are very sensitive to
algorithm to avoid obstacles and reach to the goal position.
road conditions, and thus their operational performances are
more or less limited, compared to conventional wheels.
II. KINEMATIC MODELING
On the other hand, the active caster wheel is not sensitive
to road conditions and also is able to overcome a sort of steps A. Mobility Analysis
encountered in uneven floors by using the active driving Mobility is known as the number of minimum input
parameters required to specify all the locations of the system
E.-J. Jung is with School of Electrical Engineering and Computer Science, relative to another. Grübler’s formula describing mobility is
Hanyang University, Korea. (e-mail : [email protected]) given [15] by
H. Y. Lee is with School of Electrical Engineering and Computer Science, J
Hanyang University, Korea. (e-mail : [email protected]) M = N ( L − 1) − ∑ ( N − Fi ) , (1)
J. H. LEE is with Ubiquitous Functions Research Group, Intelligent i =1
Systems Research Institute, AIST, Japan. (e-mail : [email protected])
B.-J. Yi is with School of Electrical Engineering and Computer Science,
where N is the dimensions of the allowable motion space
Hanyang University, Korea. (e-mail : [email protected]) spanned by all joints, L is the link number, J is the joint
W.K. Kim is with Department of Control and Instrumentation Engineering, number and Fi is the motion degree of freedom of the i -th
Korea University, Korea (e-mail : [email protected])
S. Y is with Institute of Engineering Mechanics and Systems, University joint. Consider a three-wheeled mobile robot shown in Fig.
of Tsukuba, Japan ([email protected])

978-1-4244-1647-9/08/$25.00 ©2008 IEEE. 1659

1(a). of the wheel and the length of the steering link, respectively.
The linear velocities at the center of the wheels are given as
v oi = θi ( sin ϕi i + cos ϕi j) × rk i = 1, 2, 3 . (3)
The linear velocity at C of the mobile robot can be
described as
JJJJJG JJJJG
v c = v o1 + η1k × O1 A1 + ω k × A1C
= θ ( sin ϕ i + cos ϕ j) × rk
1 1 1

+ η1k × ( − d cos ϕ1i + d sin ϕ1 j) , (4)

⎛l ⎞
+ ω k × ⎜ i + aj ⎟
⎝ 2 ⎠
Fig. 1. Modeling of Omni-directional mobile robot JJJJJG JJJJG
v c = v o 2 + η 2 k × O2 A2 + ωk × A2 C

We assume that every wheel retains point contact at the = θ ( sin ϕ i + cos ϕ j) × rk
2 2 2
ground, and that the wheel does not slip in the horizontal
direction, but it is allowed to rotate about the vertical axis. + η2 k × ( − d cos ϕ 2 i + d sin ϕ2 j) , (5)
Each wheel mechanism possesses its own driving and ⎛ l ⎞
steering actuators. Then, the instantaneous motion of each + ωk × ⎜ − i + aj ⎟
wheel mechanism can be modeled as two revolute joints and ⎝ 2 ⎠
one prismatic joint, as shown in Fig.1(b). Mobility of the and
JJJJJG JJJG
J
omni-directional mobile robot can be calculated as below. v c = v o 3 + η3k × O3 A3 + ωk × A3C
9
M = 3(8 − 1) − ∑ (3 − 1) = 3 (2) = θ ( sin ϕ i + cos ϕ j) × rk
3 3 3
, (6)
i =1
+ η3k × ( −d cos ϕ3 i + d sin ϕ3 j)
B. First-order Kinematics + ωk × ( -bj)
Consider the mobile robot depicted in Fig. 2. This system
where ω representing the angular velocity of the mobile
consists of three wheels, three offset link, and a top platform.
platform can be described as
Assume that the motion of the mobile robot is constrained to
ω = ηi + ϕi i = 1, 2,3 . (7)
the plane and there exists no sliding and skidding friction, but
T
that rotation of the wheel about the axis vertical to the ground v c = ⎡⎣vcx vcy ⎤⎦ of (4)-(6) and ω of (7) can be expressed as
is allowed. XYZ represents the global reference frame, and
one matrix form given by
xyz denotes a local coordinate frame attached to the mobile
platform; i, j, and k are the unit vectors of the xyz coordinate ⎡ −d sin ϕ1 − a r cos ϕ1 −a ⎤ ⎡η1 ⎤
frame. C denotes the origin of the local coordinate. u = ⎢ −d cos ϕ1 + l 2 − r sin ϕ1 l 2 ⎥ ⎢θ1 ⎥ , (8)
⎢ ⎥⎢ ⎥
l ⎢⎣ 1 0 1 ⎥⎦ ⎢⎣ϕ1 ⎥⎦
Wheel #3 A3
⎡ − d sin ϕ2 − a r cos ϕ2 − a ⎤ ⎡η2 ⎤
ϕ3
r O3 u = ⎢ −d cos ϕ 2 − l 2 −r sin ϕ 2 − l 2 ⎥⎥ ⎢⎢θ2 ⎥⎥ ,
⎢ (9)
b ⎢⎣
ĵ 1 0 1 ⎥⎦ ⎢⎣ϕ2 ⎥⎦
Y C Vc and
iˆ
ω a ϕ2 ⎡ −d sin ϕ3 + b r cos ϕ3 b ⎤ ⎡η3 ⎤
A1 u = ⎢⎢ − d cos ϕ3 −r sin ϕ3 0 ⎥⎥ ⎢⎢θ3 ⎥⎥ , (10)
X ϕ1 A2
Wheel #2

⎣⎢ 1 0 1 ⎦⎥ ⎣⎢ϕ3 ⎦⎥
d θ1 kˆwheel O2 where the operational velocity vector is defined as
O1 ⎡ vcx ⎤
Wheel #1
u = ⎢⎢ vcy ⎥⎥ . (11)
η1
Fig. 2. Kinematics description of the omni-directional mobile robot ⎢⎣ ω ⎥⎦
From (8) through (10), the mobile robot can be visualized as a
We define θ as the rotating angle of the wheel and ϕ as the parallel mechanism. The variable (η) interfacing the wheel to
steering angle between the steering link and the local x-axis. η the ground enables this model. Thus, the kinematics of the
denotes the angular displacement of the wheel relative to the mobile robot is instantaneously equivalent to that of typical
X-axis of the global reference frame. r and d denote the radius parallel robot that is connected to a fixed ground.

1660
 The intermediate coordinate transfer method [14], which posture of the mobile robot is given by the ( 3 × 1 ) vector.
was popularly employed in parallel robot community, will be ⎡ x⎤
ξ = ⎢⎢ y ⎥⎥ .
employed to derive the forward kinematic relation. Taking
(15)
the inverses of (8)-(10), we have
⎢⎣φ ⎥⎦
⎡ l ⎤
⎢ − r sin ϕ1 − r cos ϕ1 2
r cos ϕ1 − ar sin ϕ1 ⎥
⎡η1 ⎤ ⎢ ⎥
⎢θ ⎥ = 1 ⎢ d cos ϕ −d sin ϕ l
d sin ϕ + ad cos ϕ ⎥ u
⎢ 1 ⎥ dr ⎢ 1 1
2
1 1
⎥
⎢⎣ϕ1 ⎥⎦ ⎢ ⎥
⎢ r sin ϕ l ⎥
r cos ϕ dr + ar sin ϕ − r cos ϕ
⎣⎢ ⎦⎥
1 1 1 1
2
(12)
⎡ l ⎤
⎢ −r sin ϕ 2 −r cos ϕ 2 2
r cos ϕ 2 − ar sin ϕ 2 ⎥
⎡η2 ⎤ ⎢ ⎥
⎢θ ⎥ = 1 ⎢ d cos ϕ l
−d sin ϕ2 − d sin ϕ 2 + ad cos ϕ 2 ⎥ u
⎢ 2 ⎥ dr ⎢ 2
2 ⎥
⎢⎣ϕ2 ⎥⎦ ⎢ ⎥
⎢ r sin ϕ l ⎥
r cos ϕ dr + ar sin ϕ + r cos ϕ
⎢⎣ 2 2 2
2
2
⎥⎦
(13)
and Fig. 3. The omni-directional mobile robot with active caster wheels
⎡η3 ⎤ ⎡ − r sin ϕ3 − r cos ϕ3 br sin ϕ3 ⎤
⎢θ ⎥ = 1 ⎢ d cos ϕ − d sin ϕ3 −bd cos ϕ3 ⎥⎥ u . (14)
⎢ 3 ⎥ dr ⎢ 3

⎢⎣ϕ3 ⎦⎥ ⎢⎣ r sin ϕ3 r cos ϕ3 dr − br sin ϕ3 ⎦⎥ ĵ

C
iˆ
Since the omni-directional mobile robot has mobility 3, we ω
can define three minimum (or independent) coordinates. A1
However, only three inputs are not enough to avoid A1 ϕ
1
singularity. Campion, et al.[1] and Yi and Kim[12] addressed d θ1
that at least, two active wheels (i.e., two driving and two
steering motors) are necessary. In our system, we employ η1
Encoder
three active wheels (six actuators). Thus, the system has force
redundancy. In the equations (11) through (13), θ1 ,θ 2 , θ3 and Fig. 4. Time-varying parameters determined by encoders
ϕ1 , ϕ 2 , ϕ3 are the active joint variables. For a given
operational velocity vector u , the system tries to follow the l
calculated joint angular velocities in real time. YG b
YC
III. ANALYSIS OF ODOMETER
XC
C a
A. Geometric Analysis y
φ A2 kˆwheel
b ϕ2 θ2
For navigation, mobile robots should know where they are.
d
One of fundamental methods for this issue is using its η2
A1 φ O2
odometer that can be calculated geometrically.
Fig. 3 shows the omni-directional mobile robot that was ϕ1
d θ1
designed as an experimental platform. Each serial chain has kˆwheel
3-DOF in the kinematic model, but it has only two encoders
O1 η1
which measure ϕi and θ i as shown in Fig. 4. In this r
mechanism, ϕi and θ i are motor-driven joints, and ηi is a G x XG
passive joint. Fig. 5. Posture of the omni-directional mobile robot
The posture of the mobile robot can be described in terms
of the two coordinates x and y of the origin C of the moving The posture (i.e., position and orientation with respect to
frame and the orientation angle φ of the moving frame, both the global reference frame) of the mobile robot can be
obtained by the encoder information ( ϕi and θ i ) of the two
with respect to the global frame with origin at G. Hence, the

1661
chains as described in Fig.5. The rotation of the caster wheel mobile robot. Fig. 7 shows the positions of the vertical plate
with respect to the base frame, which is represented by the in the laser range finder frame during the path following. Fig.
angle ηi , can be expressed as 8 shows external measurement of the vertical plate, which is
transformed to the global frame by (21). The horizontal bar
ηi = ϕi − φ .
denotes the shape of the vertical plate.
(16)
YG
Then if φ is given, then the robot posture can be determined
YC
as follows.
XC
Step 1. Set the initial positions ( O1 , O2 ) of the wheels
C φ
and angles of the caster wheels. y
O1 : ( xO1 , yO1 ) O2 : ( xO 2 , yO 2 )
Vertical
ηi [0] = ϕi [0] = θi [0] = 0 Plate
xO1[0] = d − l / 2, yO1[0] = − a .
xO 2 [0] = d + l / 2, yO 2 [0] = −a
(17)

Step 2. Update the positions of the caster wheels

xO1[n + 1] = xO1[n] + r Δθ cos(η1[n]) x URG Sensor XG
yO1[n + 1] = yO1[n] − r Δθ sin(η1[n]) Fig. 6. External measurement system using a laser range finder
xO 2 [n + 1] = xO 2 [n] + r Δθ cos(η 2 [n])
TABLE I
yO 2 [n + 1] = yO 2 [n] − r Δθ sin(η 2 [n]) SPECIFICATION OF THE LASER RANGE FINDER
(18) Model URG-04LX(HOKUYO Co.)
Measurement Range 20~4000 mm
Step 3. Calculate the positions of the steering joints Measurement Resolution 1 mm
A1 : ( x1 , y1 ) A2 : ( x2 , y2 ) Angular Range 240 deg
x1 = xO1 − d cos(η1 ), y1 = yO1 + d sin(η1 ) Angular Resolution 0.3515625 deg

x2 = xO 2 − d cos(η 2 ), y2 = yO 2 + d sin(η 2 )
Raw Data of Laser Range Finder
(19) 1800

1600
Step 4. Obtain the posture of the mobile robot
1400
C : ( xc , yc )
⎛π ⎞ 1200
xc = x1 + b cos ⎜ + φ ⎟
Y Position (mm)

⎝6 ⎠ 1000

⎛π ⎞ 800
yc = y1 + b sin ⎜ + φ ⎟ (20)
⎝ 6 ⎠ 600

φ = atan 2 ( ( y2 − y1 ), ( x2 − x1 ) ) 400
The odometry information of the omni-directional mobile
200
robot with active caster wheels can be obtained by using this
procedure at every encoder sampling time iteratively. 0
-500 0 500 1000
X Position (mm)
B. Experiment
Fig. 7. Positions of the vertical plate in the laser range finder frame
In order to evaluate the performance of the proposed
method for odometer, an experiment is conducted for a
rectangular trajectory. In experiment, we measure the real
position of the mobile robot by using a laser range finder that
is fixed at a certain position of the global frame. Fig. 6
displays the external measurement system, and the
specification of the laser range finder is shown in Table I. The
two coordinates x and y of the origin C of the moving frame
are the center position of the vertical plate on the top of

1662
 Transformed Data of Laser Range Finder Trajectory Property
1200 1200

1000 1000

800
800
Y Position (mm)

Y Position (mm)
600
600 Trajectory
Internal Odometer
400 External Measurement
400

200
200

0
0

-200
-1200 -1000 -800 -600 -400 -200 0 200
X Position (mm)
-200
-1200 -1000 -800 -600 -400 -200 0 200
X Position (mm)
Fig. 8. Positions of the vertical plate in the global frame
Fig. 9. Trajectory property of the omni-directional mobile robot.
⎡ x ⎤ ⎡cos(θ ) − sin(θ ) ⎤ ⎡ xL ⎤ ⎡ Δx ⎤
⎢ ⎥= ⎢ ⎥⎢ ⎥+ ⎢ ⎥ , (21)
⎣ y ⎦ ⎣ sin(θ ) cos(θ ) ⎦ ⎣ yL ⎦ ⎣ Δy ⎦
Trajectory Property
50
where θ denotes the rotation of the laser sensor frame with 40
Trajectory
Internal Odometer
respect to the global frame, and xL and yL denote External Measurement
30
coordinates in the laser sensor frame, and Δx and Δy denote 20
the position of the laser sensor in the global coordinate frame.
Y Position (mm)

10
Fig. 9 shows the trajectory following property of the
0
omni-directional mobile robot. In the result of rectangular
trajectory tracking control, the solid line denotes the given -10

trajectory, the dotted line denotes the position history of the -20

mobile robot measured by odometer inside the robot, and the -30
dashed line denotes the position history measured by a URG -40
laser sensor from outside. The external measurement depicted
-50
in Fig. 9 is the collection of the center points of the horizontal -50 -40 -30 -20 -10 0 10 20 30 40 50
X Position (mm)
bars of Fig. 8. Although there are some noises owing to
resolution of the LRF sensor, it is proved that the Fig. 10. Odometry error of the omni-directional mobile robot.
omni-directional mobile robot has a good tracking
performance of the given rectangular trajectory. Fig. 10
shows the odometry error near the global origin position after IV. NAVIGATION
the path following. 1cm error occurs after traveling 4 m . All A. Navigation Algorithm
forms of mobile robots cannot help having such odometry For navigation, mobile robots should know their postures
errors. The mobile robot we developed has more complicated with respect to the global reference frame as well as the
structure compared with differential wheel type, which means positions of obstacles. Because the mobile robot we
that it can have more complicated systematic odometry error. developed has 3-DOF, more dexterous motion to avoid
Correction of systematic odometry errors in the obstacle is possible as compared to the differential type
omni-directional mobile robot is a future work. mobile robot. The first goal of navigation will be moving to
the goal point, and then the second goal will be avoiding
obstacles. To satisfy these purposes, positions of the mobile
robot, obstacles, and goal point should be estimated. The
navigation algorithm ought to minimize the distance between
the current position of the mobile robot and the goal point,
and maximizes the distances between the current position of
the mobile robot and obstacles. The performance indices are
set as follows.

1663
 min{PG = ( x − xG ) 2 + ( y − yG ) 2 } distance.
(22)
max{Pi = ( x − xi ) + ( y − yi ) },
2 2
Relationship between Ki gaine and the obstacle distance (k=10,h=700)
10
where x and y denote the positions of the mobile robot,
9
xG and yG denote the positions of the goal point, and
8
xi and yi denote the positions of the i-th obstacle. 7

Here, we would like to model the motion of the

Weight factor K i
6

omni-directional mobile robot as a kinematically redundant 5

robot. Initially, we set up the angular velocity of the mobile 4
platform as the operational velocity as follows. 3
⎡ vcx ⎤ 2

u = ω = J ⎢⎢ vcy ⎥⎥ , (23) 1

⎢⎣ ω ⎥⎦ 0
0 100 200 300 400 500 600 700 800 900 1000
Distance (mm)
J = [ 0 0 1] and
T
where ⎡⎣ vcx vcy ω ⎤⎦ denotes the
Fig. 11. Weight factor K i property with respect to the distance.
velocity of the mobile platform with respect to the global
reference frame. The general solution for (23) is given by
⎡ vcx ⎤ B. Experiment
⎢ v ⎥ = J + u + I − J + J ∇H ,
⎢ cy ⎥ ([ ] ) (24) Experimentation has been performed to verify the
navigation performance of the developed mobile robot. In the
⎢⎣ ω ⎥⎦
experiment, the odometry information was used for locating
where H defined in the homogeneous solution is the the mobile robot, and a laser range finder was used for
potential function that needs to be optimized. When the detecting obstacles and map building.
gradient of H is given as follows Fig. 12 shows the experimental environment. The primary
(
∇H = − K G
∂PG T
∂θ
+ ∑ Ki
∂Pi T
∂θ ), (25) task is that the mobile robot starts at (0, 0) and moves to the
goal point at (1000, 5000). While the mobile robot is
it implies that PG and Pi are to be optimized. Depending upon navigating, it gathers the geometric information of the
the sign of the coefficient K G and K i , the mobile robot is environment and the obstacles through a laser range finder.
controlled in such a way to minimize or maximize the given The results, the environment map and the trace of the mobile
robot, are shown in Fig. 13. The curved line in the center of
performance index. In this example, K G should be positive
the map denotes the trace that the robot has been on, and the
to minimize the distance between the robot and the goal, and dots denote the environment.
K i should be negative to maximize the distance between the In usual differential-driven mobile robots, the orientation
robot and the obstacles. Here, the vector θ = [ x y φ] of the mobile robot should be changed so that they avoid
T

obstacles. This propagates additional error of the laser range

denotes the posture variables of the mobile robot. When the
finder due to the orientation change of the mobile robot. This
angular velocity of the mobile robot is given, the
problem can be removed by using omni-directional mobile
homogeneous solutions make the mobile robot navigate.
robot, where the robot as well as the laser range finder is able
Decision of K G and K i is made as follows. K G depends on
to maintain its orientation, and thus the laser range finder is
the distance and thus it is determined by experiment. K i is able to reduce the map building error. This point is one of
determined by the following equation. merits of the omni-directional mobile robot in comparison to
ln(0.01)
Si a usual differential-driven mobile robot.
K i = ke h
, (26) The attached video clip demonstrates the navigation of the
where k denotes weighting factor, h denotes permissible omni-directional mobile robot under fixed obstacles and
approach range, and Si denotes the distance between the moving obstacles.
mobile robot and the i-th obstacle. K i is equal to 0.01× k at
Si = h according to (26). K i should vary with respect to the
distance from an obstacle to the robot, because obstacles
outside of the permissible range are not needed to be cared,
and the nearer obstacles should generate more input for the
mobile robot. Fig. 11 shows that K i varies with respect to the

1664
 [2] P. F. Muir and C. P. Newman, “Kinematic modeling of wheeled mobile
robot,” Journal of Robotics System, vol. 4, no. 2, pp. 281–340, 1987.
[3] S. K. Saha and J. Angeles, “Kinematic and dynamics of a three-wheeled
Goal 2-DOF AVG,” in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and
point Systems, pp. 1572-1577, 1989.
[4] P. Chen, S. Mitsutake, T. Isoda, and T. Shi, “Omni-directional robot
and adaptive control method for off-road running,” IEEE Trans. on
Robotics and Automation, vol. 18, no. 2, pp. 251-256, 2002.
[5] K. Tadakuma, S. Hirose, and R. Tadakuma, “Development of
VmaxCarrier2: omni-directional mobile robot with function of
step-climbing,” in Proc. IEEE Int. Conf. on Robotics and Automation,
pp. 3111-3118, 2004.
[6] S. Ziaie-Rad, et al., “A practical approach to control and
Starting self-localization of Persia omni directional mobile robot,” in Proc.
point IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, pp. 3473-3479,
2005.
[7] J. E. M. Salih, et al., “Designing omni-directional mobile robot with
Fig. 12. The picture of real environment mecanum wheel,” American Journal of Applied Sciences, vol. 3, no.5,
pp. 1831-1835, 2006.
Map building and Moving path [8] K. L. Moore and N. S. Flann, “A six-wheeled omnidiretional
6000
autonomous mobile robot,” IEEE Control System Magazine, vol. 20, no.
6, pp. 53-66, 2000.
[9] M. D. Berkemeier and L. Ma, “Discrete control for visual servoing the
5000 ODIS robot to parking lot lines,” in Proc. IEEE Int. Conf. on Robotics
and Automation, pp. 3149-3154, 2005.
4000 [10] M. Wada, A. Takagi, and S. Mori, “Caster drive mechanism for
Y Position (mm)

holonomic and omnidirectional mobile platforms with no over

constraint,” in Proc. IEEE Int. Conf. on Robotics and Automation, pp.
3000
1531-1538, 2000.
[11] N. Ushimi, M. Yamamoto, and A. Mohri, “Two wheels caster type
2000 odometer for omni-directional vehicles,” in Proc. IEEE Int. Conf. on
Robotics and Automation, pp. 497-502, 2003.
[12] J. H. Lee, S. Yuta, E. Koyanagi, and B.-J. Yi, “Command system and
1000
motion control for caster-type omni-directional mobile robot,” in Proc.
Environment
Moving path
IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, pp. 2714-2720,
0 2005.
-3000 -2000 -1000 0 1000 2000 3000 4000 [13] B.-J. Yi and W.K. Kim, “The Kinematics for redundantly actuated
X Position (mm)
omni-directional mobile robots,” Journal of Robotic Systems, vol. 19,
Fig. 13. The result of map building and navigation no. 6, pp. 255-267, 2002.
[14] R.A. Freeman and D. Tesar, “Dynamic modeling of serial and parallel
mechanisms/robotic systems,” Part I-Methodology, Part
II-Applications, 20th ASME Biennial Mechanism Conf., Orlando, FL,
V. CONCLUSION pp. 7-27, 1988.
[15] K.H. Hunt, Kinematic Geometry of Mechanisms, Clarendon Press,
In this paper, the navigation of an omni-directional mobile Oxford, 1990.
robot with three active caster wheels was investigated. We [16] T.B. Park, J.H. Lee, B.-J. Yi, W.K. Kim, B.J. You, and S.R. Oh,
derived the odometry information from the parallel mobile “Optimal design and actuator sizing of redundantly actuated
omni-directional mobile robots,” in Proc. IEEE Int. Conf. on Robotics
geometry. The accuracy of the odometry information was and Automation, pp. 732-737, 2002.
measured by the external measurement system and the [17] P. Zavlangas, et al., “Fuzzy Obstacle Avoidance and Navigation for
trajectory tracking performance was verified. Furthermore, Omnidirectional Mobile Robots,” ESIT 2000, pp. 375-382, September
2000, Aachen, Germany.
we proposed a new navigation algorithm that simultaneously
achieves the goal positioning and obstacle avoidance. The
feasibility of this algorithm was verified through
experimentation.

ACKNOWLEDGMENT
This work was supported by the GRRC Project of
Gyeonggi Province Government, Republic of Korea
(2007-041-0003- 001).

REFERENCES
[1] C. Campion, G. Batin, and B. D’Andrea-Novel, “Structural properties
and classification of kinematics and dynamics models of wheeled
Mobile Robot”, IEEE Trans. on Robot and Automation, vol. 4, no. 2,
pp. 281–340, 1987.

1665