Anda di halaman 1dari 13

1074 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO.

5, OCTOBER 2009
A Synchronization Approach to Trajectory Tracking
of Multiple Mobile Robots While Maintaining
Time-Varying Formations
Dong Sun, Senior Member, IEEE, Can Wang, Wen Shang, and Gang Feng, Fellow, IEEE
AbstractIn this paper, we present a synchronization approach
to trajectory tracking of multiple mobile robots while maintaining
time-varying formations. The main idea is to control each robot
to track its desired trajectory while synchronizing its motion with
those of other robots to keeprelative kinematics relationships, as re-
quired by the formation. First, we pose the formation-control prob-
lemas a synchronization control problemand identify the synchro-
nization control goal according to the formation requirement. The
formation error is measured by the position synchronization error,
which is dened based on the established robot network. Second,
we develop a synchronous controller for each robots translation to
guarantee that both position and synchronization errors approach
zero asymptotically. The rotary controller is also designed to ensure
that the robot is always oriented toward its desired position. Both
translational and rotary controls are supported by a centralized
high-level planer for task monitoring and robot global localization.
Finally, we perform simulations and experiments to demonstrate
the effectiveness of the proposed synchronization control approach
in the formation control tasks.
Index TermsFormation, multiple mobile robots, synchroni-
zation.
I. INTRODUCTION
S
TUDYon the formation control of swarms of mobile robots
has received increasing attention in recent years. Examples
of formation-control tasks include assignment of feasible forma-
tions, moving into a formation, maintenance of formation shape,
and switching between formations. In this paper, we will focus
on how to maintain the networked robots in a time-varying for-
mation while performing a group task as a whole. The relevant
applications include exploration [1], cooperative robot recon-
naissance [2] and manipulation [3], formation ight control [4],
satellite clustering [5], and control of groups of unmanned ve-
hicles [6], [7].
Numerous approaches to the formation control of multiple
mobile robots have been reported in the literature, such as
Manuscript received January 21, 2009; revised May 5, 2009 and July 2, 2009.
First published August 4, 2009; current version published October 9, 2009. This
paper was recommended for publication by Associate Editor G. Antonelli and
Editor L. Parker upon evaluation of the reviewers comments. This work was
supported in part by the Research Grants Council of the Hong Kong Special
Administrative Region, China, under Grant CityU 119907 and in part by the
City University of Hong Kong under Grant 7002461.
The authors are with the Department of Manufacturing Engineering and
Engineering Management, City University of Hong Kong, Kowloon, Hong Kong
(e-mail: medsun@cityu.edu.hk; cwang2@cityu.edu.hk; wenshang@cityu.edu.
hk; megfeng@cityu.edu.hk).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TRO.2009.2027384
behavior-based control, virtual structure approach, and leader-
following strategy, to name a few. In the behavior-based con-
trol [2], [8][11], several desired behaviors are prescribed for
each agent, and the nal control is derived from a weighting of
the relative importance of each behavior. With this method, it
might be difcult to describe the dynamics of the group and guar-
antee the stability of the whole system [12]. In the virtual struc-
ture approach [13][16], the entire formation is treated as a sin-
gle entity. The desired motion is assigned to the virtual structure
that traces out the trajectory for each member of the formation
to follow. However, the controller is not in decentralized archi-
tecture and may encounter difculty in some applications [12].
With the leader-following strategy [3], [12], [17][20], some
robots are designed as leaders, while others are designed as fol-
lowers. This strategy is easily implemented by using two con-
trollers only and is suitable to describe the formation of robots,
but it is hard to take into account the functioning capabilities
of different robots, i.e., ability gap of a robot [12]. There also
exist some other approaches, such as articial-potential-based
methods [21], [22] and graph-theory-based methods [23][29].
A recent approach to convex optimization strategies for coordi-
nating large-scale robot formations was reported in [30], where
shape transitions were discussed for mobile robot teams, albeit
from a high-level planner perspective.
In this paper, we propose to use a synchronization control
strategy to address the multirobot formation control problem,
thus utilizing the concept of cross-coupling approach [31]. Our
basic idea, which was rst reported in [32], is that a team of
mobile robots track each individuals desired trajectory while
synchronizing motions among the robots to keep relative kine-
matics relationship for maintaining the desired, and perhaps
time-varying, formation. A relevant work was reported in [33],
where a group of mobile agents stays connected while achiev-
ing some performance objective. Traditionally, the control loop
of each robot receives only local feedback from the controlled
robot and aims to achieve the desired tracking task without
responding to any other robots. With the proposed synchroniza-
tion control, the control loop of each robot receives feedback
from itself, as well as the others in trajectory tracking, which
is termed as the rst task; meanwhile, it takes care of the other
robots to meet the formation requirement, which is termed as the
second task. In other words, under the synchronization control,
not only the convergence of the position errors to zero but how
these position errors converge to zero will also be considered at
the same time. A pure position control without synchronization
may succeed in the rst task of the trajectory tracking but, in
1552-3098/$26.00 2009 IEEE

SUN et al.: SYNCHRONIZATION APPROACH TO TRAJECTORY TRACKING OF MULTIPLE MOBILE ROBOTS 1075
general, cannot guarantee accomplishment of the second task
of formation maintenance during the motion, due to the lack
of effort in synchronizing the robots motions. A pure position
control can only achieve the desired formation when the posi-
tion errors converge to zero but not during the whole motion.
In many applications, the robots are required to maintain the
desired formation during the whole motion period rather than
at the nal time only. The proposed synchronization control
strategy is ideally suited to the formation tasks of underlying
trajectory controls of multiple mobile robots while following
desired time-varying formations.
The cross-coupling control technology provides advantages
and opportunities to design such a synchronized controller. Over
the past decades, the cross-coupling concept has been widely
used in multiaxis motion applications, such as reducing con-
touring errors of computer numerically controlled (CNC) ma-
chines [34][39]. The concept was also incorporated into adap-
tive control architecture to solve position synchronization of
multiple axes [40], [41]. The cross-coupling technology has
been used in robotics, such as controls of mobile robots [42]
and robot manipulators [43]. Very recently, a combined synchro-
nization and tracking control law was reported in [44] and [45]
for multiple Lagrangian systems in a general setup that includes
complex congurations and different coupling links, such as di-
rectional couplings. To avoid the usage of the system dynamic
models, model-free cross-coupling controllers were introduced
in [46] and [47]. The effort to examine the stability and robust-
ness of the cross-coupled control system was reported in [48].
This paper has made the following contributions.
First, we successfully extend the synchronization approach to
formation control applications and pose the formation control
problem as a motion synchronization problem. As in [43], a
synchronization control goal is determined rst, based on the
formation requirement in the established robot network. Then,
we divide the motion control of each robot into two parts: One
is to drive the robot along the desired trajectory to achieve the
tracking control goal, which is dened as the rst task, and
the other is to synchronize each robots motion with that of
two nearby robots to achieve the synchronization goal, which
is dened as the second task. To measure the synchronicity of
the networked robots, the concept of the synchronization error is
introduced as in [41] and [43], which is dened as the differential
position error between every pair of two neighboring robots.
Second, we develop a decentralized cross-coupled controller
for each robot to achieve its position tracking while synchro-
nizing its motion with those of other robots for the desired
formation. The control algorithm utilizes feedback of both posi-
tion and synchronization errors, requires the information of the
two neighboring robots only, and responds to all linked robots
via the group network. It is proven that the proposed controller
can guarantee asymptotic convergence to zero of both position
and synchronization errors.
Third, simulations and experiments are performed on swarms
of mobile robots to demonstrate the effectiveness of the pro-
posed approach. A generalized superellipse with varying pa-
rameters is utilized to present a range of curves and, therefore,
establish a useful mathematic model to dene time-varying for-
Fig. 1. Group of two-wheel mobile robots.
mations. The result for a skewed superellipse formation was
also reported in [49].
The advantages of applying our proposed synchronization
approach to the formation control are threefold. First, the syn-
chronization control goal is determined based on the desired
formation and is then divided into a number of subgoals for
each individual robot, without discrimination in job assignment.
The potential of robots capability is not overlooked. Also, the
formation strategy can be well prescribed. Second, a synchro-
nization controller can be constructed to guarantee asymptotic
convergence to zero of both position tracking and formation
errors in a decentralized architecture for time-varying forma-
tions. The controller consists of two parallel actions for both
trajectory tracking and formation, and the weighting of the two
actions is adjustable via tuning the control parameter. Third, the
synchronization controller can be simplied by synchronizing
the motion of each robot with that of two neighboring robots. In
other words, the control of each robot requires the information
of two nearby robots only.
II. MULTIROBOT FORMATION VIA SYNCHRONIZATION
Fig. 1 illustrates a group of two-wheel mobile robots, where
q
i
= [ x
i
y
i
]
T
denotes the position coordinate of the ith robot
in xy plane, and
i
denotes the heading angle. To simplify the
analysis, we assume that the center of the mass of each robot
locates at the geometrical center of the robot. In this way, the
centripetal and Coriolis effects are not considered in the robot
dynamics, and the robot is further simplied as a point mass
robot with the following dynamics:
M
i
q
i
=
qi
, I
i

i
=
i
(1)
where M
i
and I
i
denote the inertia of the robot with xed terms,
and
qi
and
i
are torque control inputs with respect to q
i
and

i
, respectively.
Consider the control problem of guiding and positioning a
group of nmobile robots, as shown in Fig. 1, along the boundary
(curve) of a 2-D compact set. In a similar manner to [50], we
introduce a time-varying desired shape for each robot, which
is denoted by S(, t), where denotes a 2-D position vector
and t the time. The boundary of S(, t) is parameterized by a
2-D planar curve, which is denoted by S(, t) = 0. Assign the
target position q
d
i
to the ith robot, and q
d
i
must be located on

1076 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO. 5, OCTOBER 2009
the curve such that S(q
d
i
, t) = 0. Our objective is to determine
the appropriate control inputs for dynamics (1) such that the
ith robot converges to its target position q
d
i
while maintaining
its position in the desired shape S(, t). Note that although
only the position q
i
is explicitly controlled in the formation,
the orientation
i
affects the formation implicitly. In this study,
the desired heading
d
i
of the ith robot is dened such that the
robot is always oriented toward the robots desired position q
d
i
.

d
i
keeps updated during the motion. When the robot is exactly
located at its desired position, the desired heading
d
i
is the same
as the actual heading
i
, and no update is needed.
Dene the position and heading errors of the ith robot as
e
i
= q
d
i
q
i
and
i
=
d
i

i
, respectively. Besides the tra-
ditional robot control goal of e
i
0 and
i
0 as time
t , the robots here are also required to achieve a formation
control goal of maintaining on the desired curve, which can be
formulated as q
i
, S(q
i
, t) = 0, where i = 1, . . . , n.
In this paper, we propose to utilize the synchronization control
concept to solve the formation control problem. The basic idea
of the synchronization control is to regulate the motions of the
robots while they track the desired positions q
d
i
to ensure the
robots to maintain in the required boundary (curve) such that
S(q
i
, t) = 0. Note that a pure position control for q
i
q
d
i
cannot guarantee the formation control goal to be achieved,
because the curve S(q
i
, t) = 0 is dependent on not only the
position q
i
but on the relative relationship of q
i
as well, with
respect to the others at time t.
To measure the synchronicity of the robots, we introduce the
concept of the position synchronization error, which arises from
the synchronization constraint, as required by the formation
goal. The following example will showhowthis synchronization
constraint is determined based on the formation control goal
S(q
i
, t) = 0. Another example was reported in [51].
ExampleConsider that n robots are required to maintain in
an ellipse curve during the motions. The coordinate q
i
of the
ith robot is required to meet the following constraint:
S(q
i
, t) = 0 :
_
q
i
(t) =
_
x
i
(t)
y
i
(t)
_
=
_
cos
i
(t)
sin
i
(t)
_ _
a(t)
b(t)
_
= A
i
(t)
_
a(t)
b(t)
__
(2)
where a and b denote the longest and the shortest radii of
the ellipse, respectively, and
i
= tanh (b sin
i
/a cos
i
), with

i
= tanh [y
i
/x
i
], denotes the angle of the robot lying on the
ellipse with respect to the center of the ellipse. Assume that the
robots are not located in the longest or the shortest axis of the
ellipse such that the inverse of A
i
exists. Then, the synchroniza-
tion constraint to q
i
can be derived as follows:
A
1
1
q
1
= A
1
2
q
2
= = A
1
n
q
n
=
_
a
b
_
. (3)
Fromthe aforementioned example, it is seen that the synchro-
nization constraint can be generally represented in the form
c
1
q
1
= c
2
q
2
= = c
n
q
n
(4)
where c
i
denotes the coupling parameter of the ith robot, and its
inverse exists based on (3). Further, (4) holds at all the desired
coordinates q
d
i
, namely
c
1
q
d
1
= c
2
q
d
2
= = c
n
q
d
n
. (5)
Subtracting (4) from (5) yields the following synchronization
goal:
c
1
e
1
= c
2
e
2
= = c
n
e
n
. (6)
Implicitly, (6) represents the formation control goal, which
can be further divided into n subgoals of c
i
e
i
= c
i+1
e
i+1
. Note
that when i = n, we denote n+1 as 1.
Then, the position synchronization errors can be dened as
a subset of all possible pairs of two neighboring robots in the
following way [52]:

1
= c
1
e
1
c
2
e
2

2
= c
2
e
2
c
3
e
3
.
.
.

n
= c
n
e
n
c
1
e
1
(7)
where
i
denotes the synchronization error of the ith robot. Ob-
viously, if the synchronization error
i
= 0 for all i = 1, . . . , n,
the synchronization goal (6) is achieved automatically.
It is worth noting that the synchronization error is used to mea-
sure the formation effect and is not equivalent to the position-
tracking error. For systems having closed-loop chain structures,
employment of the synchronization error provides each robot
with motion information both from itself and from the other
robots, and hence, the motions of all robots in the group are
coordinated.
Remark 1: The topology of the networked robots is designed
in terms of the robots physical positions in the group when
performing the group task. A principle that may be followed is
that the two physically close robots are coded as two neighbors
for easy sensorial connection. As a result, all robots in one group
are linked, either directly or indirectly, as a whole. The necessary
requirement for the applied shape S(, t) is that the shape can
be represented mathematically such that the synchronization
constraint (4) can be given.
Remark 2: In some formations, the two neighborly coded
robots may be spatially far and, hence, not sensorially con-
nected. This usually happens when the formation shape is not
geometrically closed, e.g., in a line formation, the rst robot
(robot 1) and the last robot (robot n) are far apart but coded
as two neighbors. In case there exists a sensorial problem be-
tween these two robots, we can utilize a high-level planner (as
seen in Fig. 2) to globally localize the two robots such that
the two robots can know the information between each other.
Alternatively, we can also simply remove the direct synchro-
nization request between these two robots, since the two robots
synchronize with the other robots, respectively, and the whole
group of the robots are still linked as a whole. Removal of the
synchronization between the two robots can be mathematically
achieved by setting the synchronization error between these two

SUN et al.: SYNCHRONIZATION APPROACH TO TRAJECTORY TRACKING OF MULTIPLE MOBILE ROBOTS 1077
Fig. 2. Block diagram of the overall control architecture.
robots to zero. In the relevant work [33], sensorial connection
was analytically studied.
Now, the control problem becomes to drive both the position
error e
i
and the synchronization error
i
to zero. The ith robot
can be designed to approach its desired position q
d
i
while syn-
chronizing its motion with those of its two connected robots
i 1 and i + 1. In this way, the control of each robot does not
require the information of all robots except for its neighbors,
and hence, the implementation is simplied.
III. CONTROL DESIGN
Fig. 2 illustrates a block diagram of a two-level control ar-
chitecture used for the formation control. In level 1, there is
a centralized high-level planer for task command generation,
task monitoring (such as detecting malfunctioned robots and
responding), and global localization of the robots. In level 2,
there are decentralized servo controls of the robots and senso-
rial connection among them. Between the two levels is a data
server. Obviously, these two levels have different functions in
accomplishing complex formation control tasks.
This section will focus on the decentralized servo controls of
the robots in level 2. A synchronous tracking controller for the
robots translation is developed. The robots orientation is con-
trolled to be always oriented toward the robots desired position.
The issue of how the robot group responds to the malfunctioned
robot will also be discussed briey at the end of this section.
Without loss of generality, we assume that each robot can obtain
the position information of its two nearby robots, via global or
local localization in the two-level control frame, as shown in
Fig. 2. The relevant works on localization (such as ours [53])
have been reported in the literatures.
A. Synchronous Formation Controller
To make the position and synchronization errors e
i
and
i
both converge to zero, we dene a coupled position error E
i
that links these two errors into one equation, i.e.,
E
i
= c
i
e
i
+
_
t
0
(
i

i1
) d (8)
where is a diagonal positive gain matrix. As is seen from (7)
and (8), this coupled position error for robot i feeds back the
information of the two neighboring robots i 1 and i + 1. Note
that when i = 1, we denote i 1 as n.
Differentiating E
i
with respect to time yields

E
i
= c
i
e
i
+ c
i
e
i
+ (
i

i1
) . (9)
To achieve E
i
0 and

E
i
0, we introduce a command
vector u
i
that leads to a combined position and velocity error,
which is expressed as follows:
u
i
= c
i
q
d
i
+ c
i
e
i
+ (
i

i1
) + E
i
(10)
where is a diagonal positive gain matrix. Denition of u
i
in
(10) leads to the following position/velocity vectors:
r
i
= u
i
c
i
q
i
= c
i
e
i
+ c
i
e
i
+ (
i

i1
) + E
i
=

E
i
+ E
i
. (11)
We then design a controller to drive r
i
to zero, such that
the coupled errors E
i
and

E
i
tend to zero as well. For easy
implementation, such a controller is ideally designed in a de-
centralized architecture, thus considering the synchronization
between each robot and its two neighbors only.
We nally design a torque input for controlling the robots
translation as follows:

qi
=M
i
c
1
i
( u
i
c
i
q
i
) +K
ri
c
1
i
r
i
+c
T
i
K

(
i

i1
) (12)
where K
ri
and K

are positive feedback control gains. The last


term in (12) is used to compensate for the effect due to addition
of the cross-coupling control to the overall system dynamics.
The necessity of introducing this term will be shown in the
stability analysis.
Substituting (12) into the robot translational dynamics in (1)
yields the following closed-loop dynamics:
M
i
c
1
i
r
i
+ K
ri
c
1
i
r
i
+ c
T
i
K

(
i

i1
) = 0. (13)
Theorem 1: The proposed synchronous controller (12) leads
to asymptotic convergence of both the position tracking and syn-
chronization errors to zero, namely, e
i
0 and
i
0 as time
t , under the conditions that the control gain K
ri
is large
enough to satisfy
min
(K
ri
)
max
_
M
i
(d/dt)
_
c
1
i
_
c
i
_
,
where
min
() and
max
() denote the minimum and maximum
eigenvalues of the matrices, respectively.
Proof: Dene a Lyapunov function candidate as
V =
n

i=1
_
1
2
_
c
1
i
r
i
_
T
M
i
c
1
i
r
i
+
1
2

T
i
K

i
_
+
n

i=1
1
2
__
t
0
(
i

i1
) d
_
T
K

_
t
0
(
i

i1
) d.
(14)
Differentiating V with respect to time yields

V =
n

i=1
_
_
c
1
i
r
i
_
T
M
i
c
1
i
r
i
+
_
c
1
i
r
i
_
T
M
i
d
dt
_
c
1
i
_
r
i
+
T
i
K


i
_
+
n

i=1
(
i

i1
)
T
K

_
t
0
(
i

i1
) d. (15)

1078 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO. 5, OCTOBER 2009
Multiplying both sides of (13) by
_
c
1
i
r
i
_
T
yields
_
c
1
i
r
i
_
T
M
i
c
1
i
r
i
+
_
c
1
i
r
i
_
T
K
ri
c
1
i
r
i
+r
T
i
K

(
i

i1
)
= 0. (16)
Substituting (16) into (15) yields

V =
n

i=1
_
_
c
1
i
r
i
_
T
_
K
ri
M
i
d
dt
_
c
1
i
_
c
i
_
c
1
i
r
i
_

i=1
_
r
T
i
K

(
i

i1
)

+
n

i=1

T
i
K


i
+
n

i=1
(
i

i1
)
T
K

_
t
0
(
i

i1
) d. (17)
We now analyze the term

n
i=1
r
T
i
K

(
i

i1
) in (17). It
directly follows that
n

i=1
r
T
i
K

(
i

i1
)
= r
T
1
K

1
r
T
1
K

n
+r
T
2
K

2
r
T
2
K

1
+ r
T
n
K

n
r
T
n
K

n1
= r
T
1
K

1
r
T
2
K

1
+r
T
2
K

2
r
T
3
K

2
+ r
T
n
K

n
r
T
1
K

n
=
n

i=1
(r
i
r
i+1
)
T
K

i
. (18)
Utilizing (7)(9) and (11), we have
r
i
r
i+1
=

E
i
+ E
i


E
i+1
E
i+1
= c
i
e
i
+c
i
e
i
c
i+1
e
i+1
c
i+1
e
i+1
+ (2
i

i1

i+1
)
+ (c
i
e
i
c
i+1
e
i+1
) +
_
t
0
(2
i

i1

i+1
) d
=
i
+ (2
i

i1

i+1
) +
i
+
_
t
0
(2
i

i1

i+1
) d. (19)
Then, substituting (19) into (18) yields
n

i=1
r
T
i
K

(
i

i1
)
=
n

i=1

T
i
K

i
+
n

i=1

T
i
K

i
+
n

i=1
(
i

i+1
)
T
K

(
i

i+1
)
+
n

i=1
(
i

i+1
)
T
K

_
t
0
(
i

i1
) d. (20)
In (20), we utilize the result that
n

i=1
(2
i

i1

i+1
)
T

i
=
n

i=1
(
i

i+1
)
T
(
i

i+1
).
Finally, substituting (20) into (17) and utilizing the condition
of Theorem 1, we have

V
n

i=1
_
_
c
1
i
r
i
_
T
_

min
(K
r
)

max
_
M
i
d
dt
_
c
1
i
_
c
i
__
c
1
i
r
i
_

i=1

T
i
K

i

n

i=1
_

i

i+1
_
T
K

(
i

i+1
)
0. (21)
It is thus concluded that the dened Lyapunov function (14)
has negative semidenite time derivative. From (21), we know
that r
i
and
i
are bounded. From (13), we know that r
i
is
bounded. From (19), we know that
i
is bounded. Therefore, r
i
and
i
are uniformly continuous since r
i
and
i
are bounded.
Then, from Barbalats lemma, r
i
0 and
i
0 as time t
. The synchronization goal (6) is achieved.
We now prove e
i
= 0 when r
i
= 0 and
i
= 0. From (11),
r
i
= 0 implies E
i
= 0 when t . Combining all equations
in (8) from 1 to n, we have
c
1
e
1
+ c
2
e
2
+ + c
n
e
n
= 0. (22)
Substituting (6) into (22) yields c
1
e
1
= c
2
e
2
= =
c
n
e
n
= 0. Since the inverse of c
i
exists, e
i
= 0. Therefore,
Theorem 1 is proved.
It appears that the control parameters in (8) and K

in (12)
dominate the control of the synchronization error. As seen in
(8), the value of determines the weight of the synchroniza-
tion error
i
in the coupled position error E
i
and, thus, affects
the synchronization effort in the whole control action. As
increases, the synchronization will be enhanced. In practical
applications, should be chosen by taking a balance between
position and synchronization controls. The parameter K

, ac-
companied by
i

i1
in (12), ensures the stability of the
system when adding the cross-coupling control with to the
overall system dynamics. It will be further discussed how to
get the desired boundedness of the synchronization error by
adjusting K

in Section III-B.
To control the robots heading, we propose to utilize a general
computed torque approach to design of the control input
i
, i.e.,

i
= I
i
_

d
i
+ k
vi

i
+ k
pi

i
_
(23)
where k
vi
and k
pi
are computed torque control gains. The de-
sired heading
d
i
, as said before, is dened such that the robot is
always oriented toward its desired position. In other words, the
orientation control serves to achieve the robots translation.
Substituting (23) into the robot rotational dynamics in (1)
yields the closed-loop dynamics as follows:

i
+ k
vi

i
+ k
pi

i
= 0. (24)

SUN et al.: SYNCHRONIZATION APPROACH TO TRAJECTORY TRACKING OF MULTIPLE MOBILE ROBOTS 1079
This directly yields
i
= 0 and

i
= 0 as time t
and guarantees the stability properties of the rotation.
Remark 3: If robot i malfunctions, e.g., lags much behind
on its trajectory, the other robots will respond under the syn-
chronization control. The centralized high-level planar in level
1 will monitor the task and make the decision whether robot i is
still suitable to remain in the team. The decision making will be
based on the facts whether the robot loses capabilities to execute
the task (i.e., drive the position error to zero) and communicate
with the other robots. If robot i has to be abandoned by the
team, the other robots will treat robot i as a virtual robot and
assume its position error to be zero in the follow-on motion.
In this way, the remaining robots can still follow the previous
network topology.
B. Discussions
Two issues are further discussed in this section. One is the
boundedness on the formation error. Another is the robustness
achieved by adaptive synchronization control.
1) Boundedness: We nowanalyze the boundedness through-
out the transient response. Given a upper bound on the formation
error, in the following, we will discuss the lower bound on the
control parameter that achieves the desired upper bound on the
error.
Since the system is asymptotically stable, as concluded from
Theorem 1, given an upper bound
i
on
i
, there exists a nite
time T > 0 such that after t > T, one has
i
(t)
i
. In
addition, at t = T
V (T) =
n

i=1
_
1
2
_
c
1
i
(T)r
i
(T)
_
T
M
i
c
1
i
(T)r
i
(T)
_
+
n

i=1
_
1
2

T
i
(T)K

i
(T)
_
+
n

i=1
1
2
__
T
0
(
i

i1
) d
_
T
K

_
T
0
(
i

i1
) d (25)
which is bounded. If we dene the rst term on the right-hand
side of (25) as A, it then follows from (25) that
V (T) A
=
n

i=1
_
1
2

T
i
(T)K

i
(T)
_
+
n

i=1
1
2
__
T
0
(
i

i1
) d
_
T
K

_
T
0
(
i

i1
) d

i=1
_
1
2
K

2
_
+
n

i=1
_
1
2
K

(2
i
T)
2
_
= K

_
1
2
+ 2 T
2
_
n

i=1
_

2
_
. (26)
Finally, the lower bound for K

can be determined as
K


V (T) A
_
1
2
+ 2 T
2
_
n
i=1
(
i

2
)
. (27)
2) Adaptive Control for Robustness: In the controller (12),
the robot inertia M
i
is assumed to be known. In practice, the
inertia M
i
may be uncertain. Dene

M
i
(t) as the estimate of
the robot inertia M
i
. An adaptive control can be designed as

qi
=

M
i
(t)c
1
i
( u
i
c
i
q
i
) + K
ri
c
1
i
r
i
+ c
T
i
K

(
i

i1
).
(28)
The estimated inertia

M
i
(t) is subject to the adaptation law

M
i
(t) =
i
c
1
i
( u
i
c
i
q
i
)c
1
i
r
i
(29)
where
i
is a diagonal positive-denite control gain. Dene the
estimation error as

M
i
(t) = M
i


M
i
(t). (30)
Then, the adaptive control law can be rewritten as

M
i
(t) =
i
c
1
i
( u
i
c
i
q
i
)c
1
i
r
i
. (31)
Substituting the controller (28) into the robot dynamics model
leads to the following closed-loop dynamics:
M
i
c
1
i
r
i
+ K
ri
c
1
i
r
i
+ c
T
i
K

(
i

i1
)
=

M
i
(t)
_
c
1
i
( u
i
c
i
q
i
)
_
. (32)
In a similar manner to Theorem 1 and [43], we can prove that
the aforementioned adaptive control law leads to asymptotic
convergence to zero of both the position tracking and synchro-
nization errors.
The aforementioned analysis indicates that an adaptive con-
trol can be well incorporated into the developed synchronization
scheme. Although only inertia uncertainty is discussed here, the
method can be extended to the cases with more complex mod-
eling uncertainty and/or some external disturbances.
IV. SIMULATIONS
Simulations were performed rst to verify the effectiveness
of the proposed synchronization control approach. It is assumed
that all the required formations are composed of regular closed,
smooth, and simple planar curves. A generalized superellipse
with varying parameters, as seen later, is used to represent dif-
ferent kinds of formation curves

x
i
a

2/m
+

y
i
b

2/m
= 1 (33)
which can also be represented as
_
x
i
= a cos
m

i
y
i
= b sin
m

i
(34)
where m denotes the exponent index that may be time-varying,
and a, b, and
i
have been dened in (2). In the simulations,

i
is xed, and its value can be known at the beginning. Based

1080 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO. 5, OCTOBER 2009
Fig. 3. Switch from an ellipse to a rounded rectangle.
on
i
, each robot can be indexed. Besides the exponent m, the
variation of radii a and b also affects the curve signicantly. A
study on switch between two different ellipses with exchanged
radii a and b was reported in [32].
With different exponent m, (33) and (34) represent a range
of shapes including rectangles, ovals, ellipses, and diamonds, in
categories of hyperellipses (m < 1) and hypoellipses (m > 1).
The following simulation study will consider a switch from an
ellipse (m
0
= 1) to a rounded rectangle (m
f
= 1/8), as seen in
Fig. 3, where a and b are xed.
In the simulation, there are in total 20 mobile robots, which
are denoted by little squares in Fig. 3, all located on an ellipse
curve at the beginning time. During the switch, all the robots
are required to maintain in a desired time-varying hyperellipse
curve, with the exponent index changed in the following way:
m(t) = m
0
+ (m
f
m
0
)
_
1 e
t
_
(35)
where m
0
= 1 and m
f
= 1/8 denote the initial and nal values
of the exponent index. All the 20 robots have different inertia
M
i
, which is expressed by M
i
= diag{0.5 + 0.075i, 0.5 +
0.075i}, where i = 1, . . . , 20.
The desired trajectory of the ith robot is designed according
to the required formation task (34) and is given as
q
d
i
(t) =
_
x
d
i
(t)
y
d
i
(t)
_
=
_
cos
m(t)

i
sin
m(t)

i
_ _
a
b
_
= A
i
(t)
_
a
b
_
. (36)
The coupling parameter matrix was then dened as
c
i
(t) = A
1
i
(t) =
_
cos
m(t)

i
sin
m(t)

i
_
1
. (37)
The sampling period was set to 0.005 s in the simulation.
Two control algorithms were applied in the simulation for
comparison purpose: One was the proposed synchronous con-
Fig. 4. Formation switch froman ellipse to a rounded rectangle and trajectories
of robots, with synchronous control.
trol (12), and the other was the nonsynchronous control by re-
moving all kinds of coupling among the robots in the controller,
which was equivalent to a standard feedforward plus a feedback
control. The control parameters of the synchronous controller
(12) were chosen as = diag{15, 15}, = diag{35, 35},
K

= diag{10, 10}, and K


ri
= diag{120, 120}. To implement
the nonsynchronous control, the control parameters were chosen
as = diag{0, 0}, = diag{35, 35}, K

= diag{0, 0}, and


K
ri
= diag{120, 120}. Note that the same coupling parameter
matrix c
i
was used to calculate the synchronization error in both
cases for a fair comparison.
Fig. 4 illustrates the actual formation shapes and the trajec-
tories of the robots in the switch from the initial ellipse to the
nal rounded rectangle under the synchronous control. Fig. 5(a)
and (b) illustrates the position and synchronization errors in x-
and y-direction of the 20 robots under the proposed synchronous
control. Both the position and synchronization errors growfrom
zero up to some nonzero values and, subsequently, decrease and
converge tozerouponreachingthe nal desiredformation. Fig. 6
illustrates the results of the nonsynchronous control. It is seen
that the robots have large transient synchronization errors under
nonsynchronous control, which degrades the performance of
the formation. With the proposed synchronous control, the syn-
chronization errors are greatly reduced, and therefore, a better
formation can be achieved.
We further study how the robot group responds to a particular
robot that is observed to malfunction. Assume that robot 18 was
found to lag much behind its desired trajectory soon after the
task started. Due to the use of synchronization control, all the
other robots had to slow down to wait for this robot to catch the
team. After recognizing that robot 18 could not resume working
properly, the group decided to abandon this robot at 0.75 s. The
remaining 19 robots then continued the movement and treated
robot 18 as a virtual robot with an assumption that its position
error was zero. Fig. 7(a) and (b) illustrates the position and
synchronization errors of the case. It is seen that due to failure
of robot 18, the position and synchronization errors of all the

SUN et al.: SYNCHRONIZATION APPROACH TO TRAJECTORY TRACKING OF MULTIPLE MOBILE ROBOTS 1081
Fig. 5. Results of synchronous control. (a) Position errors. (b) Synchronization
errors.
robots in the group increased signicantly until robot 18 was
abandoned after 0.75 s.
V. EXPERIMENTS
We further carried out experiments on a group of three mobile
robots to verify the proposed synchronous control method. The
three robots are P3DX mobile robots, as shown in Fig. 8. The
control inputs
qi
and

i
were obtained by the inputs acting
on the left and the right driving wheels, which are denoted by

li
and
ri
, as given in the Appendix. The experiments were
performed in two cases.
A. Case 1: Triangle Formation
In this case, the three robots were controlled to maintain a se-
ries of equilateral triangles. The three robots were located at the
three vertices of the equilateral triangle. Dene a circumcircle
that connects the three vertices of the triangle. The center of the
circumcircle locates at the geometrical central point of the tri-
angle. The switch among the equilateral triangles was modeled
as the switch among the circumcircles with time-varying radius
and orientation, as shown in Fig. 9. Taking reference of (2), the
coordinate q
i
of the ith robot, where i = 1, . . . , 3, was subject
Fig. 6. Results of nonsynchronous control. (a) Position errors. (b) Synchro-
nization errors.
to the following constraint:
S(q
i
, t) = 0 :
_
q
i
(t) =
_
x
i
(t)
y
i
(t)
_
=
_
cos
i
(t)
sin
i
(t)
_ _
R(t)
R(t)
_
= A
i
(t)
_
R(t)
R(t)
__
(38)
where R(t) is the time-varying radius of the circumcircle, and

i
(t) was dened in (2). Since a = b = R(t),
i
(t) =
i
.
In the experiment, both the size and the orientation of the
triangle were changed by varying R(t) and
i
(t), respectively.
R(t) was scheduled to change in the following way:
R(t) = R
0
+ (R
f
R
0
)
t
t + e
1t
(39)
where R
0
and R
f
denote the initial and the nal radii of the
circumcircle and were chosen as R
0
= 2.4 m and R
f
= 4.8 m,
respectively.
i
(t) was scheduled to change as

i
(t) =
i0
+ (
if

i0
)
t
t + e
1t
(40)
where
i0
and
if
are the initial and the nal values for the
ith robot and were chosen as
10
= 0

,
1f
= 120

,
20
= 0

2f
= 120

, and
30
= 0

,
3f
= 0

. After system calibra-


tion, the inertia of each robot was estimated as 1.8 kgm
2
.

1082 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO. 5, OCTOBER 2009
Fig. 7. Results when dealing with the failed robot. (a) Position errors.
(b) Synchronization errors.
Fig. 8. Three mobile robots in the experiment.
Dene the coupling parameter matrix as
c
i
(t) = A
1
i
(t) =
_
cos
i
(t)
sin
i
(t)
_
1
.
Fig. 9. Switch among triangle formations.
The sampling time period was chosen as 200 ms. The control
parameters used in the synchronous controller are
= diag{1, 1}, = diag{2, 2}, K

= diag{5, 5}
K
r1
= diag{5, 5}, K
r2
= diag{5, 5}, K
r3
= diag{5, 5}.
As in the simulations, two control algorithms of the pro-
posed synchronous control (12) and the nonsynchronous con-
trol were implemented, respectively. To implement the non-
synchronous controller, the coupling among the robots was re-
moved, and the parameters were chosen as = diag{0, 0} and
K

= diag{0, 0}, and and K


ri
were the same as in the syn-
chronous case. Note that the same coupling parameter c
i
was
used to calculate the synchronization error for both the syn-
chronous and nonsynchronous cases. The robots were localized
based on their initial positions and the moving distances mea-
sured by the motor encoders.
The right-hand side of Fig. 9 illustrates the legend of the tri-
angle formation change in the experiment. Fig. 10 illustrates
the position and synchronization errors of the three robots in
x- and y-direction, respectively, under the proposed synchro-
nization control (12). Fig. 11 illustrates the position and syn-
chronization errors of the robots under the nonsynchronous
control. It is seen that both the control methods could ensure
good convergence of the robot position errors. The synchro-
nization errors under the synchronous controller were much
smaller than those under the nonsynchronous controller. The
synchronization errors actually represent the formation errors.
Fig. 12 illustrates the heading angle errors of the three robots
under the computed torque controller (23), where the control
gains were K
p1
= diag{10, 10}, K
p2
= diag{10, 10}, K
p3
=
diag{10, 10}, K
v1
= diag{10, 10}, K
v2
= diag{10, 10}, and
K
v3
= diag{10, 10}. Fig. 13 illustrates the control inputs
li
and
ri
on the two wheels of each robot, respectively, under the
synchronous control.
B. Case 2: Ellipse Formation
To further verify the robustness of the proposed adaptive
synchronization control (28) and (29) to the model uncertainty,
the three robots were controlled to form a part of ellipse (see
Fig. 14) and then switch among a series of ellipse formations.
At the beginning, the robots were in a straight line with the
same heading angle, as shown on the left-hand side of Fig. 15.
Then, the robots spread out to explore a series of formations

SUN et al.: SYNCHRONIZATION APPROACH TO TRAJECTORY TRACKING OF MULTIPLE MOBILE ROBOTS 1083
Fig. 10. Synchronous control results. (a) Position errors. (b) Synchronization
errors.
that are all part of ellipses. The coordinate q
i
of the ith robot is
subject to (2). The longest and the shortest radii of the ellipse
were changed as
a(t) = a
0
+ (a
f
a
0
)
t
t + e
1t
(41)
b(t) = b
0
+ (b
f
b
0
)
t
t + e
1t
(42)
where a
0
= 3 m and a
f
= 5 m are the initial and the nal de-
sired longest radii of the ellipse, and b
0
= 1 m and b
f
= 3 m
are the initial and the nal desired shortest radii of the ellipse.
In the experiment, let
i
[as dened in (2)] change to

i
(t) =
i0
+ (
if

i0
)
t
t + e
1t
(43)
where we chose
10
= 30

,
1f
= 75

,
20
= 0

,
2f
= 25

,
and
30
= 30

,
3f
= 75

. All inertias of the robots were


estimated to be zero at the beginning. Dene the same cou-
pling coefcient as in case 1. The control gains were chosen as
Fig. 11. Nonsynchronous control results. (a) Position errors. (b) Synchroniza-
tion errors.
Fig. 12. Heading angle errors of the three robots.

1084 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO. 5, OCTOBER 2009
Fig. 13. Control torques.
Fig. 14. Three robots in an ellipse.
Fig. 15. Switch among formations.
Fig. 16. Adaptive synchronous control results. (a) Position errors. (b) Syn-
chronization errors.
follows:
= diag{1, 1}, = diag{1.3, 1.3}, K

= diag{1, 1}
K
r1
= diag{10, 10}, K
r2
= diag{10, 10}
K
r3
= diag{10, 10},
i
= diag{0.1, 0.1}.
The sampling time period was 200 ms.
The right-hand side of Fig. 15 illustrates the legend of the
ellipse formation change in the experiment. Fig. 16 illustrates
good convergence of the position and synchronization errors
of the three robots under the proposed adaptive synchronous
control, although all inertias of the robots were assumed to be
unknown. These results demonstrate the validity of incorpora-
tion of adaptive control into synchronization approach to solve
the model uncertainty problem.

SUN et al.: SYNCHRONIZATION APPROACH TO TRAJECTORY TRACKING OF MULTIPLE MOBILE ROBOTS 1085
VI. CONCLUSION
This paper presents a synchronization control approach to
controlling swarms of mobile robots to track the desired trajec-
tories while synchronizing the motions among them to maintain
relative kinematics relationships, as required by the formation.
The formation control problem is successfully posed as a syn-
chronization control problem. The concept of the position syn-
chronization error, which is dened as differential position error
between every pair of two neighboring robots, is introduced to
measure the performance of the formation. The proposed syn-
chronous controller guarantees asymptotic convergence to zero
of both position and synchronization errors of each robot in
translation. A rotary controller drives the robot to always be
oriented toward its desired position. Both simulation and ex-
perimental studies are nally performed to demonstrate the ef-
fectiveness of the proposed approach. The future work includes
incorporation of the robot nonholonomic dynamics into the syn-
chronization control, topology design of the networked robots,
and a study on robustness to the environmental uncertainties.
APPENDIX
Dene the torque inputs acting on the left and the right driving
wheels of the ith robot as
li
and
ri
. The transformation matrix
from (
li
,
ri
) to (
qi
,
i
) is expressed as
B
i
(
i
) =
1
d
_

_
cos
i
cos
i
sin
i
sin
i
R R
_

_ (44)
where d and R have been shown in Fig. 1. Based on B
i
(
i
),
qi
and
i
can be derived as

qi
=
_

qi
(x)

qi
(y)
_
=
_

_
cos
r
(
li
+
ri
)
sin
r
(
li
+
ri
)
_

_
(45)

i
=
R
r
(
li

ri
). (46)
The inverse relationship from (
qi
,
i
) to (
li
,
ri
) can be
obtained by

li
=
r
2
_

qi
(x)
cos
+

i
R
_

ri
=
r
2
_

qi
(x)
cos


i
R
_
(47)
or

li
=
r
2
_

qi
(y)
sin
+

i
R
_

ri
=
r
2
_

qi
(y)
sin


i
R
_
. (48)
To avoid singularity, (47) should be used when
i
is around
zero or 180

, and (48) should be used when


i
is around 90

or
270

.
REFERENCES
[1] D. Fox, W. Burgard, H. Kruppa, and S. Thrun, A probabilistic approach
to collaborative multi-robot localization, Auton. Robots, vol. 8, no. 3,
pp. 325344, Jun. 2000.
[2] T. Balch and R. Arkin, Behavior-based formation control for multi-robot
systems, IEEE Trans. Robot. Autom., vol. 14, no. 6, pp. 926939, Dec.
1998.
[3] H. G. Tanner, S. G. Loizou, and K. J. Kyriakopoulos, Nonholonomic
navigation and control of multiple mobile manipulators, IEEE Trans.
Robot. Autom., vol. 19, no. 1, pp. 5364, Feb. 2003.
[4] M. Mesbahi and F. Hadaegh, Formation ying of multiple spacecraft via
graphs, matrix inequalities, and switching, AIAA J. Guid., Control, Dyn.,
vol. 24, pp. 369377, Mar. 2001.
[5] C. R. McInnes, Autonomous ring formation for a planar constellation of
satellites, AIAA J. Guid., Control, Dyn., vol. 18, no. 5, pp. 12151217,
1995.
[6] F. Giulietti, L. Pollini, and M. Innocenti, Autonomous formation ight,
IEEE Control Syst. Mag., vol. 20, no. 6, pp. 3444, Jun. 2000.
[7] D. J. Stilwell and B. E. Bishop, Platoons of underwater vehicles, IEEE
Control Syst. Mag., vol. 20, no. 6, pp. 4552, Dec. 2000.
[8] J. R. T. Lawton, R. W. Beard, and B. J. Young, A decentralized approach
to formation maneuvers, IEEE Trans. Robot. Autom., vol. 19, no. 6,
pp. 933941, Dec. 2003.
[9] L. E. Parker, ALLIANCE: An architecture for fault tolerant multirobot
cooperation, IEEE Trans. Robot. Autom., vol. 14, no. 2, pp. 220240,
Apr. 1998.
[10] S. Berman, Y. Edan, and M. Hamshidi, Navigation of decentralized au-
tonomous automatic guided vehicles in material handling, IEEE Trans.
Robot. Autom., vol. 19, no. 4, pp. 743749, Aug. 2003.
[11] M. Long, A. Gage, R. Murphy, and K. Valavanis, Application of the
distributed eld robot architecture to a simulated deming task, in Proc.
IEEE Int. Conf. Robot. Autom., Barcelona, Spain, Apr. 2005, pp. 3204
3211.
[12] H. Takahashi, H. Nishi, and K. Ohnishi, Autonomous decentralized con-
trol for formation of multiple mobile robots considering ability of robot,
IEEE Trans. Ind. Electron., vol. 51, no. 6, pp. 12721279, Dec. 2004.
[13] M. A. Lewis and K. H. Tan, High precision formation control of mobile
robots using virtual structures, Auton. Robot., vol. 4, pp. 387403, 1997.
[14] M. Egerstedt and X. Hu, Formation constrained multi-agent control,
IEEE Trans. Robot. Autom., vol. 17, no. 6, pp. 947951, Dec. 2001.
[15] R. W. Beard, H. Lawton, and F. Y. Hadaegh, A coordination architecture
for spacecraft formation control, IEEE Trans. Control Syst. Technol.,
vol. 9, no. 6, pp. 777790, Nov. 2001.
[16] W. Kang, N. Xi, and A. Sparks, Formation control of autonomous agents
in 3D workspace, in Proc. IEEE Int. Conf. Robot. Autom., San Francisco,
CA, Apr. 2000, pp. 17551760.
[17] J. P. Deasi, V. Kumar, and P. Ostrowski, Modeling and control of for-
mations of nonholonomic mobile robots, IEEE Trans. Robot. Autom.,
vol. 17, no. 6, pp. 905908, Dec. 2001.
[18] A. K. Das, R. Fierro, V. Kumar, J. P. Ostrowski, J. Spletzer, and C. J.
Taylor, A vision-based formation control framework, IEEE Trans.
Robot. Autom., vol. 18, no. 5, pp. 813825, Oct. 2002.
[19] J. Huang, S. M. Farritor, A. Qadi, and S. Goddard, Localization and
follow-the-leader control of a heterogeneous group of mobile robots,
IEEE/ASME Trans. Mechatron., vol. 11, no. 2, pp. 205215, Apr. 2006.
[20] J. Chen, D. Sun, J. Yang, and H. Chen, A leader-follower formation
control of multiple nonholonomic mobile robots incorporating receding-
horizon scheme, Int. J. Robot. Res., vol. 28, 2009.
[21] P. Ogren, E. Fiorelli, and N. E. Leonard, Cooperative control of mobile
sensor networks: Adaptive gradient climbing in a distributed environ-
ment, IEEE Trans. Autom. Control, vol. 40, no. 8, pp. 12921302, Aug.
2004.
[22] R. Sepulchre, D. Paley, and N. E. Leonard, Stabilization of planar col-
lective motion: All-to-all communication, IEEE Trans. Autom. Control,
vol. 52, no. 5, pp. 811824, May 2007.
[23] J. A. Fax and R. M. Murray, Information ow and cooperative control
of vehicle formations, IEEE Trans. Autom. Control, vol. 49, no. 9,
pp. 14651476, Sep. 2004.
[24] A. Jadbabaie, J. Lin, and A. S. Morse, Coordination of groups of mobile
autonomous agents using nearest neighbor rules, IEEE Trans. Autom.
Control, vol. 48, no. 9, pp. 9881001, Sep. 2003.
[25] L. Moreau, Stability of multiagent systems with time-dependent commu-
nication links, IEEE Trans. Autom. Control, vol. 50, no. 2, pp. 169182,
Feb. 2005.

1086 IEEE TRANSACTIONS ON ROBOTICS, VOL. 25, NO. 5, OCTOBER 2009
[26] R. Olfati-Saber and R. M. Murray, Consensus problems in networks of
agents with switching topology and time-delays, IEEE Trans. Autom.
Control, vol. 49, no. 9, pp. 101115, Sep. 2004.
[27] R. Olfati-Saber, Flocking for multi-agent dynamic systems: Algorithms
and Theory, IEEE Trans. Autom. Control, vol. 51, no. 3, pp. 401420,
Mar. 2006.
[28] W. Ren and R. W. Beard, Consensus seeking in multi-agent systems
under dynamically changing interaction topologies, IEEE Trans. Autom.
Control, vol. 50, no. 5, pp. 655661, May 2005.
[29] C. Belta and V. Kumar, Optimal motion generation for groups of robots:
A geometric approach, ASME J. Mech. Des., vol. 126, pp. 6370, 2004.
[30] J. C. Derenick and J. R. Spletzer, Convex optimization strategies for
coordinating large-scale robot formations, IEEE Trans. Robot., vol. 23,
no. 6, pp. 12521259, Dec. 2007.
[31] Y. Koren, Cross-coupled biaxial computer controls for manufacturing
systems, ASME J. Dyn. Syst., Meas., Control, vol. 102, pp. 265272,
1980.
[32] D. Sun and C. Wang, Controlling swarms of mobile robots for switching
between formations using synchronization concept, in Proc. IEEE Int.
Conf. Robot. Autom., Rome, Italy, Apr. 2007, pp. 23002305.
[33] M. Ji and M. Egerstedt, Distributed coordination control of multi-agent
systems while preserving connectedness, IEEE Trans. Robot., vol. 23,
no. 4, pp. 693703, Aug. 2007.
[34] A. Rodriguez-Angeles and H. Nijmeijer, Mutual synchronization of
robots via estimated state feedback: Acooperative approach, IEEETrans.
Control Syst. Technol., vol. 12, no. 4, pp. 542554, Jul. 2004.
[35] Q. Zhong, Y. Shi, J. Mo, and S. Huang, A linear cross coupled control
system for high speed machining, Int. J. Adv. Manuf. Technol., vol. 19,
pp. 558563, 2002.
[36] S. S. Yeh and P. L. Hsu, Estimation of the contouring error vector for the
cross-coupled control design, IEEE/ASME Trans. Mechatron., vol. 7,
no. 1, pp. 4451, Mar. 2002.
[37] M. T. Yan, M. H. Lee, and P. L. Yen, Theory and application of a combined
self-tuning adaptive control and cross coupling control in a retrot milling
machine, Mechatron., vol. 15, no. 2, pp. 193211, Mar. 2005.
[38] T. C. Chiu and M. Tomizuka, Contouring control of machine tool feed
drive systems: A task coordinate frame approach, IEEE Trans. Control
Syst. Technol., vol. 9, no. 1, pp. 130139, Jan. 2001.
[39] D. Sun and M. C. Tung, Asynchronization approach for the minimization
of contouring errors of CNC machine tools, IEEE Trans. Autom. Sci.
Eng., vol. 6, no. 4, Oct. 2009.
[40] M. Tomizuka, J. S. Hu, and T. C. Chiu, Synchronization of two motion
control axes under adaptive feedforward control, ASME J. Dyn. Syst.,
Meas., Control, vol. 114, no. 2, pp. 196203, 1992.
[41] D. Sun, Position synchronization of multiple motion axes with adaptive
coupling control, Automatica, vol. 39, no. 6, pp. 9971005, Jun. 2003.
[42] L. Feng, Y. Koren, and J. Borenstein, Cross-coupling motion controller
for mobile robots, IEEE Control Syst., vol. 13, no. 6, pp. 3543, Dec.
1993.
[43] D. Sun and J. K. Mills, Adaptive synchronized control for coordination
of multi-robot assembly tasks, IEEE Trans. Robot. Autom., vol. 18, no. 4,
pp. 498510, Aug. 2002.
[44] S. J Chung and J. J. E. Slotine, Cooperative robot control and concurrent
synchronization of Lagrangian systems, IEEE Trans. Robot., vol. 25,
no. 3, pp. 686700, Jun. 2009.
[45] S. J. Chung, U. Ahsun, and J. J. E. Slotine, Application of synchronization
to formation ying spacecraft: Lagrangian approach, AIAA J. Guid.,
Control Dyn., vol. 32, no. 2, pp. 512526, Sep. 2008.
[46] Y. Koren and C. C. Lo, Variable-gain cross-coupling controller for con-
touring, Ann. CIRP, vol. 40, no. 1, pp. 371374, 1991.
[47] D. Sun, X. Y. Shao, and G. Feng, A model-free cross-coupled control for
position synchronization of multi-axis motions: Theory and Experiments,
IEEETrans. Control Syst. Technol., vol. 15, no. 2, pp. 306314, Mar. 2007.
[48] S. Yeh and P. Hsu, Analysis and design of integrated control for multi-
axis motion systems, IEEE Trans. Control Syst. Technol., vol. 11, no. 3,
pp. 375382, May 2003.
[49] D. A. Paley, N. E. Leonard, and R. Sepulchre, Stabilization of symmetric
formations to motion around convex loops, Syst. Control Lett., vol. 57,
pp. 209215, 2008.
[50] M. Y. Hsieh and V. Kumar, Pattern generation with multiple robots, in
Proc. IEEE Int. Conf. Robot. Autom., May 2006, pp. 24422447.
[51] Y. Su, D. Sun, L. Ren, and J. K. Mills, Integration of saturated PI syn-
chronous control and PD feedback for control of parallel manipulators,
IEEE Trans. Robot., vol. 22, no. 1, pp. 202207, Feb. 2006.
[52] D. Sun, L. Ren, J. K. Mills, and C. Wang, Synchronous tracking control
of parallel manipulators using cross-coupling approach, Int. J. Robot.
Res., vol. 25, no. 11, pp. 11371148, Nov. 2006.
[53] H. Chen, D. Sun, and J. Yang, Global localization of multirobot forma-
tions using ceiling vision SLAM strategy, Mechatron., vol. 19, no. 5,
pp. 618628, Aug. 2009.
Dong Sun (S95M00SM08) received the Bach-
elors and Masters degrees in mechatronics and
biomedical engineering from Tsinghua University,
Beijing, China, and the Ph.D. degree in robotics and
automation from the Chinese University of Hong
Kong, Hong Kong.
He was a Postdoctoral Researcher with the Uni-
versity of Toronto, Toronto, ON, Canada, where he
is currently an Adjunct Professor. He was a Research
and Development Engineer in Ontario industry. Since
2000, he has been with the Department of Manufac-
turing Engineering and Engineering Management, City University of Hong
Kong, Kowloon, Hong Kong, where he is currently a Full Professor. His current
research interests include robotics manipulation, multirobot systems, motion
controls, and biological processing automation.
Prof. Sun is a Professional Engineer in the province of Ontario. From 2004
to 2008, he was an Associate Editor of the IEEE TRANSACTIONS ON ROBOTICS.
Can Wang received the B.S. degree from the De-
partment of Electrical Engineering, Zhejiang Uni-
versity, Hangzhou, China, in 2001, the M.S. de-
gree in system-on-chip from Lund University, Lund,
Sweden, in 2005, and the Ph.D. degree with the De-
partment of Manufacturing Engineering and Engi-
neering Management, City University of Hong Kong,
Kowloon, Hong Kong, in 2009.
His current research interests include coordination
of multiple robot systems, synchronization control,
and neural networks.
Wen Shang received the B.Sc. degree in automation
from Shandong University, Jinan, China, in 1999 and
the Ph.D. degree in automatic control from Southeast
University, Nanjing, China, in 2005.
She is currently a Senior Research Associate with
Suzhou Research Center, City University of Hong
Kong, Kowloon, Hong Kong. Her current research
interests include mobile robots, multirobot learning
control, and multirobot coordination.
Gang Feng (S90M92SM95F09) received the
B.Eng. and M.Eng. degrees in automatic control
from Nanjing Aeronautical Institute, Nanjing, China,
in 1982 and 1984, respectively, and the Ph.D. de-
gree in electrical engineering from the University of
Melbourne, Melbourne, Vic., Australia, in 1992.
Since 2000, he has been with the City University
of Hong Kong, Kowloon, Hong Kong, where he is
currently a Chair Professor and an Associate Provost.
During 19921999, he was a Lecturer/Senior Lec-
turer with the School of Electrical Engineering, Uni-
versity of New South Wales, Sydney, N.S.W., Australia. He was an Associate
Editor of the Journal of Control Theory and Applications. He is an Associate
Editor of Mechatronics. His current research interests include piecewise linear
systems, robot networks, and intelligent systems and control.
Prof. Feng received an Alexander von Humboldt Fellowship during 1997
1998 and the IEEE TRANSACTIONS ON FUZZY SYSTEMS Outstanding Paper
Award in 2007. He was an Associate Editor of the IEEE TRANSACTIONS ON
SYSTEMS, MAN, AND CYBERNETICS, PART C and was a member of the Confer-
ence Editorial Board of the IEEE Control Systems Society. He is an Associate
Editor of the IEEE TRANSACTIONS ON AUTOMATIC CONTROL and the IEEE
TRANSACTIONS ON FUZZY SYSTEMS.

Anda mungkin juga menyukai