Proceedings of the 2007 IEEE/RSJ International ThA9.

3
Conference on Intelligent Robots and Systems
San Diego, CA, USA, Oct 29 - Nov 2, 2007

Self-healing for Mobile Robot Networks with Motion

Synchronization
Fei Zhang and Weidong Chen, Member, IEEE

Abstract—The objective of self-healing in mobile robot self-healing in mobile robot networks is to maintain not only
networks is to maintain not only logical topology but also logical topology but also physical one of the network to solve
physical one of a network when robots fail. An interaction the problem of energy consumption and blink zones.
dynamics model is first established to describe both logical and The self-healing of sensor networks has been studied in
physical topologies of the network. Considering the mobility of
sensor networks for many years. However, many literatures
mobile robot networks, we propose a recursive, distributed
topology control for self-healing when mobile robots fail, and were focused on wireless sensor networks without mobility.
give a metric of the topology structure for evaluating the Boonma et al. [6] proposed a biologically-inspired,
performence of recovered network topologies. Then, we prove decentralized architecture of sensor networks to self-heal
the stability of motion synchronization with the topology control false positive sensor data, with the mobility of mobile robot
based on Lyapunov exponent. Finally, the results of simulations networks ignored. Furthermore, distributed control is one of
have demonstrated the validity of the proposed modeling and
the most challenging problems of designing self-healing
control methods.
control. Zhang et al. [1] proposed a topology control for
I. INTRODUCTION self-healing and showed that the control can improve the
robustness of motion synchronization, but its control was not
M OTION synchronization of mobile robots has been
defined as follows: a large number of mobile robots
with local interacting could achieve a global objective of
completely distributed and the stability analysis of
self-healing was not provided.
Stability of motion synchronization is one of the most
synchronizing speeds of all robots [1]. It is mainly inspired by important characteristics in groups of mobile robots, and is
the research on animal behavior [2-5], for example, widely studied in the related research. Lyapunov function is
self-organizing flocks, schools and swarms in nature always the main method to analyze the stability. Gazi et al. [12]
tend to keep synchronized motions. Synchronization is also a investigated the stability properties of swarm aggregation and
fundamental concept, and is a universal phenomenon in many showed that the individual agents will form a cohesive swarm
research fields of science and technology, including sensor in a finite time. Olfati-Saber et al. [13] gave the Lyapunov
networks [6], complex dynamical networks [7] and stability of disagreement functions for consensus problems in
communication networks [8]. In multi-robotics, many networks of agents with switching topology and time-delays.
complex practical tasks and applications can be decomposed Although these conclusions about stability conditions have
into a series of motion synchronization, for example large been successfully applied in their corresponding models and
object transportation and manipulation [9], formation controls, it is necessary for other controls to search for the
generation and maintenance [10], exploration and suitable Lyapunov functions. Therefore, we expect to provide
map-building in an unknown environment [11], and a more general tool of analyzing the stability of motion
reconnaissance and surveillance for military missions. synchronization for mobile robot networks.
Self-healing is viewed as one of these complex tasks based In recent years, the research on synchronization in complex
on motion synchronization. It refers to the process of dynamical networks based on an interaction dynamics model
recovering topologies and system performances of networks has achieved many important conclusions on the stability and
from failed robots. Clearly, the disadvantages of only robustness [7, 14-16]. They have shown that synchronization
self-healing communication topologies include more energy of a complex dynamical network can be decided by the
consumption with the enlarged communicating range in network topology and the maximum Lyapunov exponent of
wireless networks, and more blind zones with the limited the individual nodes. However, the research on this topic has
sensing range in sensor networks. Different with general not involved in self-healing of mobile robot networks.
networks, mobile robot networks possess the feature of the The main contribution of this paper is to provide a
mobility of physical topologies. Therefore, the objective of theoretical modeling and control method for self-healing of
mobile robot networks with motion synchronization. We
This work was supported in part by the Natural Science Foundation of propose a fully distributed topology control for self-healing,
China under grant 60475032 and the Shu Guang Project of Shanghai. and give a metric of the topology structure to evaluate the
Fei Zhang is with Department of Automation, Shanghai Jiao Tong performance of the network topology. Based on Lyapunov
University, Shanghai, 200240, China.
Weidong Chen is with Department of Automation, Shanghai Jiao Tong
exponent, we also prove the global stability of the network
University, Shanghai, 200240, China (e-mail: [email protected]). with the proposed topology control.

1-4244-0912-8/07/$25.00 ©2007 IEEE. 3107

 The paper is organized as follows. Section II establishes an robots, the graph G representing the network is undirected,
interaction dynamics model of mobile robot network. In i.e., AT = A . Suppose that the network is connected without
Section III, we propose a recursive, distributed topology isolate clusters. Then G is connected. The neighbors of the
control for self-healing. Section IV proves the stability of the robot i are the robots having active connection with i. Let
network with the topology control. The results of simulations N e (i ) denote the index set of neighbors of i, i.e.,
in Section V have demonstrated the validity of the topology
control for self-healing. Conclusions are given in Section VI. N e (i )  { j | aij = 1} ⊆ {1, 2," , n} . (3)
Furthermore, the degree of the robot i is equal to the number
II. MODEL OF MOBILE ROBOT NETWORK of its neighbors.
Now we establish an interaction dynamics model The topologies of mobile robot networks are often
describing logical and physical topologies of a network. In described by K-neighbor models [18], where the degree of
this paper, we assume that each robot only knows the each robot is required to be equal to or less than K, i.e.,
information about its neighboring ones without any time di ≤ K , i = 1, 2," , n . For example, Fig. 1 shows four typical
delay of communication. topologies of robot networks with K=3, 4, 6, and 8,
respectively.
A. Logical Topology of Mobile Robot Network
Consider a network having n robots, whose logical B. Motion Model of Mobile Robot Network
topology can be viewed as a graph G defined by a set of Consider each robot operating in an N-dimensional space.
vertices V and a set of edges E connecting the vertices. The Let pi ∈ ℜ N and xi ∈ ℜ N with i = 1, 2," , n denote the
vertex vi ∈ V , i = 1, 2," , n , corresponds to the robot i, and position and speed of the robot i, respectively. According to
the edge ( vi , v j ) ∈ E , i, j = 1, 2," , n , corresponds to an the motion models in [1, 7, 12, 13], we propose a double
integrator system to describe the continuous-time dynamics
active communication connection between the robots i and j.
of each mobile robot:
The coupling matrix A = ( aij ) ∈ ℜn×n represents the coupling
⎧ p i = xi ,
configuration of the network [1, 7], where ⎨ (4)
⎩ xi = ui + ui , i = 1, 2," , n,
e c

⎧0, ( vi , v j ) ∉ E
⎪ where uie stands for the individual N-dimensional dynamics
aij = ⎨ , i ≠ j , i, j = 1, 2," , n , (1)
⎪⎩1, ( vi , v j ) ∈ E of the robot i, expressed as:
uie = f ( pi , xi ) . (5)
indicates which vertices are linked by an edge, and thus
which robots communicate. Moreover, For example,
n f ( pi , xi ) = am (v(t ) − xi ) , i = 1, 2," , n , (6)
aii = − ∑
j =1, j ≠ i
aij = − di , i = 1, 2," , n . (2)
where am > 0 and v(t ) represents the value of a velocity
where di denotes the degree of the robot i. field in the environment. The control uic stands for the effect
Remark: In this paper, the coupling matrix A is somewhat of the neighboring robots upon the robot i. Given a whole
different with the adjacency matrix in graph theory [17]. number k > 0 and a switched matrix A(k ) , uic can be
Because of bi-directional communication between two described as:
n n
uic = c ∑ aij (k ) ⋅ ( p j − qij ) + c ∑ aij (k ) x j , i = 1, 2," , n , (7)
j =1 j =1

where c > 0 represents the coupling strength of the network,

and qij = p dj − pid represents the desired relative position of
the robots j and i, with p dj standing for the synchronized
(a) K=3 (b) K=4
(desired) position of the robot j. In this paper, qij represents
the physical topology of the network.
The benefit of the proposed model is that it allows one can
describe not only logical topology but also physical one of a
mobile robot network.

III. TOPOLOGY CONTROL FOR SELF-HEALING

(c) K=6 (d) K=8 Self-healing represents the self-organized collective
Fig. 1 Four typical topologies of mobile robot networks with behavior of mobile robot networks when robots fail. It refers
K-neighbor model. to the process of filling the blank locations of the failed robots

3108
and recovering both logical and physical topologies from through the following rules.
robot failures. The topology control for self-healing should Rule 1: The neighbors j ∈ N e (if ) of if with d j ≤ dif
satisfy:
become the candidate of filling the failed robot, whose the
1) It can prevent the whole network from being separated
index set is denoted by:
into two or more small groups.
2) It can recover the network topology from failed robots { }
Ca (if )  j | d j ≤ dif , j ∈ N e (if ) . (9)
and keep system properties of the whole network with regard Rule 2: The robot in Ca (if ) with the smallest degree will
to the stability and robustness of motion synchronization. fill the blank location of the failed robot, denoted by is .
A. Network Topology with Failed Robots Rule 3: If there exist two or more robots with the same,
Now we study the network topology when robots fail. In [1, smallest degree, then one robot is randomly chosen to be the
19], the second-largest eigenvalue λ2 of the coupling matrix filling robot is . Furthermore, a mark M = qis if is generated
A is used to evaluate the stability and robustness of the whole and sent to the robot is′ , qis′is = M , is′ ∈ N e (is ) . In next step,
network. A small value of λ2 indicates the better stability and
the robot is′ will fill the new blank of is , if the neighbors
robustness of motion synchronization of the network [1, 7].
j ∈ Ca (is ) does not satisfy Rule 2. The mark M will also be
Moreover, λ2 = 0 , iff the network is broken into two or more
sent to the next robot until not finding the corresponding
groups.
robot.
Let Arf ∈ ℜn×n and A rf ∈ ℜ( n −[ Δn ])×( n −[ Δn ]) denote the Note that the mark is used to avoid self-healing repeatedly
coupling matrices of the original network with n robots and for one robot. Thus, it forms a recursive, distributed topology
the new network after failure of [ Δn ] , respectively. Here, control algorithm of self-healing as follows.
Self-healing Algorithm:
[ Δn] stands for the smaller but nearest integer to the real
Step 1: k = 1 . Calculate the coupling matrix A(k ) .
number Δn . Let also λ2rf and λ2rf denote the second-largest Step 2: k = k + 1 . If one robot obtains the information
eigenvalues of A and A , respectively. Moreover, assume
rf rf
about its neighbor’s failure, it will compare its degree with
that the robots i1 , i2 , " , i[ Δn ] have failed in the network. dif and judge whether it belongs to Ca (if ) .
Step 3: If there exists a robot with the smallest degree in
Then the new coupling matrix A rf from the original matrix
Ca (if ) , it is the filling robot is . Go to Step 6.
Arf can be formed according to [1, 19].
Step 4: If there exist two or more robots with the same,
Ref. [19] guaranteed that if a small fraction Δ (0 < Δ << 1)
smallest degree, the robot with the mark M in Ca (if ) will be
of robots whose degrees are sufficiently small
the filling robot is . Moreover, the robot is will send M to the
di < d 0 , i = 1, 2," ,[Δn] are deleted in the network ( n >> 1 ),
corresponding robot in its neighbors. Go to Step 6.
then
Step 5: If there exist two or more robots with the same,
λ2rf ≈ λ2rf . (8) smallest degree and without a mark, then one robot is chosen
It means that the second-largest eigenvalue of the coupling randomly in Ca (if ) with the smallest degree to be the filling
matrix is held almost fixed so that the synchronization robot is . A mark M is generated according to Rule 3.
stability of the network also remains. Assume that the degree
Step 6: The action of filling the blank location of the failed
of each robot in the K-neighbors model is far less than the size
robot is described here. The filling robot is cuts off the
of the whole network, i.e., K  n , then (8) can be also
applied in our model, because the degree of each robot connections with its origin neighbors N e (is ) , and creates new
di ≤ K , i = 1, 2," , n is sufficiently small. connections with N e (if ) . Thus, the coupling matrix A(k ) is
Ref. [1] has shown that the whole network will be broken obtained by:
into two small groups when the robots in one row or column ⎧ aij (k ) = aij (k − 1), i, j = 1, 2," , n − [Δn],
fail, that is λ2rf = 0 . Therefore, it is necessary to self-heal the ⎪a (k ) = a (k ) = 0, j ∈ N (i ),
⎪ is j jis e s
network topology from failed robots. ⎪
⎨ ais j (k ) = a jis (k ) = 1, j ∈ N e (if ), (10)
B. Self-healing Rules and Algorithm ⎪ n
⎪ a (k ) = −
Through controlling the network topology [20], the ⎪⎩ ii ∑ aij (k ), i = 1, 2," , n.
j =1, j ≠ i
network can fill the blank locations of the failed robots. Let
Other failed robots can also be substituted by (10).
A(1) = Arf denote the coupling matrix after robots fail. Step 7: If a new blank location appears, then go to Step 2.
Assume that the robot if has failed with its neighbors N e (if ) . Otherwise, self-healing finishes.
Our topology control for self-healing can be described

3109
 C. Properties of the Switched Topology with Self-healing ⎡0 0⎤
⎥ and yi = [ pi − qi xi ] ∈ ℜ2N ,
T
Denote Γ = ⎢
Now we study the properties of the switched topology with ⎣IN IN ⎦
Self-healing Algorithm. Assume that the network is i = 1, 2," , n, and rewrite (13) as:
connected after robots fail, i.e., λ2 (1) ≠ 0 . n
The aim of Self-healing Algorithm is to substitute the y i = F ( yi ) + c ∑ aij (k )Γy j , i = 1, 2," , n, (14)
j =1
filling robots is1 , is2 ,…, is[Δn] for the failed ones i1 , i2 ,…,
k = s (t ) , A(k ) ∈ Ω ,
i[ Δn] with d s j ≤ d j , j = 1, 2," , [ Δn ] . In K-neighbors models,
which is a hybrid system with a continuous-state yi and a
the process of self-healing is to move the blank locations of discrete-state A(k ) that belongs to a finite set of switched
the failed robot out of the network. Let ks denote the final
coupling matrices Ω = { A(k ), k = s (t )} . Let s (t ) : ℜ≥ 0 → Λ Ω
step number of self-healing.
Lemma: Consider the mobile robot network (4). It follows denote a switching signal where Λ Ω ⊂ ` is the index set
with Self-healing Algorithm when robots fail. Then, the associated with the elements of Ω .
following statements hold: Note that the relative positions of the neighboring robots
1) The whole network is prevented from being separated are also stable when achieving the synchronized state (12).
into two or more small groups, i.e., λ2 (ks ) ≠ 0 . Let pi′ = pi − qi , and the state pi′ is synchronized at the
2) The network topology recovers from failed robots and stable states:
the second-largest eigenvalue of the coupling matrix of the p1′(t ) = p2′ (t ) = " = pn′ (t ) = p1 (t ), t → ∞ . (15)
network is kept, i.e., λ2 (ks ) ≈ λ2 (ks − 1) ≈ " ≈ λ2 (1) . Denote S (t ) = [ p1 (t ) v(t ) ] ∈ ℜ 2 N . Thus, the synchronized
Proof: First, to prove statement 1), suppose that λ2 (kp ) = 0 states of the dynamical network (14) becomes
with kp > 0 . Because λ2 (1) ≠ 0 , it is clear that the new blank y1 (t ) = y2 (t ) = " = yn (t ) = S (t ), t → ∞ . (16)
positions causes the broken network. Then the neighbors of Assume that the network is connected without isolate
the filling robot will fill the new blank positions, and robots. Then its coupling matrix A(k ) is therefore a
therefore self-healing will continue, i.e., kp < ks . According symmetric irreducible matrix. In this case, zero is an
to Rule 1-3, the blank locations of the failed robots are moved eigenvalue of A(k ) with multiplicity 1 and all the other real
out of the network finally. Consequently, λ2 (ks ) ≠ 0 . eigenvalues of A(k ) are strictly negative [7], denoted by
Second, prove statement 2). Because the robot with the 0 = λ1 (k ) ≥ λ2 (k ) ≥ λ3 (k ) ≥ " ≥ λn (k ) , k = s (t ) . (17)
small degree substitutes the failed robot for each step, i.e.,
Denote by hmax = h1 , h2 ," , hN the corresponding
d s j ≤ d j , j = 1, 2," , [ Δn ] , based on (8) we have
Lyapunov exponents of each individual N-dimensional
λ2 (ks ) ≈ λ2 (ks − 1) ≈ " ≈ λ2 (1) . (11) mobile robot. If it is stable, the maximum Lyapunov
Consequently, Self-healing Algorithm is effective. □ exponents hmax will be negative. In the following, assume all
the individual robots are stable with hmax < 0 . Assume that
IV. STABILITY ANALYSIS
any interswitching time is enough large during switching
Hereafter, the mobile robot network (4) is said to achieve topologies.
synchronous speeds if Theorem: Consider the network (14) with identical stable
x1 (t ) = x2 (t ) = " = xn (t ) → v(t ), t → ∞ , (12) robots. It follows with Self-healing Algorithm when robots
where v(t ) ∈ ℜ N can be an equilibrium speed, a leader’s fail. Then, the synchronized states (16) are exponentially
speed or a value of a velocity field, given by (6). Now we stable.
investigate the stability condition of motion synchronization Proof: The exponential stability of (16) is transformed to
with Self-healing Algorithm. the exponential stability of the following systems [7]:
First, the model (4) can be expressed as: ω = [ DF ( s) + cλi (k )Γ]ω , i = 1, 2," , n , k = s(t ) , (18)
⎡ p i ⎤ ⎡ xi ⎤ n
⎡ 0 0 ⎤ ⎡ ( p j − qij ) ⎤ where DF ( s ) represents the Jacobian of F ( s ) about s. To
⎢ x ⎥ ⎢ f ( p , x ) ⎥ ∑ aij (k ) ⎢ I
= + c ⎥⎢ ⎥, (13) prove its stability, recall the concept of transversal Lyapunov
⎣ i⎦ ⎣ i i ⎦ j =1 ⎣ N IN ⎦ ⎣ xj ⎦
exponents [21]. For every eigenvalue λi (k ), i = 2," , n ,
i = 1, 2," , n, k = 1, 2" , ks ,
k = s (t ) , the corresponding transversal Lyapunov exponents
where I N denotes the N-dimensional identity matrix. Let
for (18) is a function of λi (k ) , denoted as L j (λi (k )) ,
qi = pid − p1d denote the desired relative distance between the
j = 1, 2," , N , and given by
robot i and 1. Note that qi just simplifying the following
L j (λi (k )) = h j + cλi (k ) , j = 1, 2," , N , k = s (t ) . (19)
analysis does not affect the distributed control for
self-healing. To stabilize the synchronized states, if suffices to require that

3110
 60 65 75

55 60 70

50 55 65
22 23 24 25 21 22 23 24 25 21
21 23 24 25
45 50 60 22

40 16 17 18 19 20 45 16 18 55
19 20 16 18 19 20

y
y

y
35 40 50
11 12 13 14 15 11 12 13 14 15 11 12 13 14 15
30 35 45
6 7 8 9 10 6 7 8 9 10 6 7 8 9 10
25 30 40

20 2 25 1 35
1 3 4 5 2 3 4 5 1 2 3 4 5
15 20 30

10 15 25
20 30 40 50 60 20 30 40 50 60 70 30 40 50 60 70 80

x x x
(a) Topology at t=1.5 (s) (b) Topology at t=2.0 (s) (c) Topology at t=3.5 (s)

115 -0.42

110 -0.44

105 23 24 25
21
-0.46
100
22 18 19 20
16 -0.48
95
y

λ2
90 12 13 14 15 -0.5
11

85 10
6 7 8 9 -0.52

80
4 5 -0.54
1 2 3
75
-0.56
70
80 90 100 110 120
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
x
Time(s)
(d) Topology at t=7.5 (s)
Fig. 3 Second-largest eigenvalue of coupling matrix in the case of
Fig. 2 Simulation results of self-healing for one failed robot. self-healing for one failed robot.

these L j (λi (k )) for each λi (k ), i = 2," , n, k = s (t ) be of the simulation were shown in Fig. 2 and 3.
In Fig. 2, the plots of (a)-(d) represent the topologies of the
negative. According to Lemma, when satisfying (17) and
mobile robot network at the time t=1.5, 2.0, 3.0 and 7.5 (s),
hmax < 0 , it indicates that the following equivalent inequality
respectively, where the black points marked by numbers and
should hold: the red arrows are the positions and speeds of the robots,
Lmax (λi (k )) = hmax + cλ2 (k ) < 0 , k = s (t ) . (20) correspondingly. Fig. 2(a) shows the original topology of the
The minimum in (20) always exists and is achieved because network. Robot 17 was failed at the time t=2.0. Because robot
Ω is a finite set. That is, the synchronized states (16) are 22 and 23 had the same, smallest degree (d=3) in the
exponentially stable with Self-healing Algorithm. □ neighbors of 17. The robot 22 was randomly chosen to fill in
the blank position of 17 in Fig. 2(b) according to Self-healing
V. SIMULATIONS Algorithm. Meanwhile, robot 21 filled in the new blank
The simulations utilized the programming of MATLAB, position of 22 in Fig. 2(c). Because there was no robot with
and performed the process of self-healing in mobile robot the smaller degree than 21, self-healing finished. Fig. 2(d)
networks. The operating space of robots was a 2-dimensional shows the final topology after self-healing.
plane ℜ 2 in simulations, i.e., N = 2 . The step length in In Fig. 3, the second-largest eigenvalue of the coupling
matrix decreased from -0.4367 to -0.5889 with the topology
simulations was 0.05 (s). The initial positions were selected
control. It means that the topology control is effective to keep
randomly within the interval [-1, 1] around the corresponding
system performance of the network fixed.
positions according to K-neighbor model with K=6, and the
initial speeds were chosen randomly with [-5, 5]. B. Self-healing for Failed Robots
A. Self-healing for One Failed Robot Now we illustrate the second simulation to demonstrate the
self-healing behavior in a robot network of 100 robots with
We performed two simulations of the proposed topology
K=6 for failed robots. In the simulation, 10 robots failed
control for self-healing in mobile robot networks when robots
randomly in the network. The total time of experiments was
fail. In the first simulation, one robot failed in a network of 25
10 seconds and other parameters were configured as above.
robots ( n = 25 ). The total time of experiments was 10 seconds
In Fig. 4, the plots of (a) and (b) represent the topologies
and other parameters were configured as above. The results

3111
 -0.09

-0.1

-0.11

-0.12

λ2
-0.13
y

y
-0.14

-0.15

-0.16

-0.17

-0.18
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

x x Time(s)
(a) Topology before self-healing (b) Topology after self-healing
Fig. 5 Second-largest eigenvalue of coupling matrix in the
Fig. 4 Simulation results of self-healing for failed robots. case of self-healing for failed robots.

before and after self-healing, respectively, where the circle [6] P. Boonma, P. Champrasert, and J. Suzuki, “BiSNET: A
biologically-inspired architecture for wireless sensor network,” In Proc.
represents the sensing range of the robot. Obviously, of 2006 IEEE Inter. Conf. on Autonomic and Autonomous Systems
self-healing has decreased the blind zones of sensing in the (ICAS ’06), July, 2006, pp. 54-60.
network. The dashed and the solid lines in Fig. 5 represent the [7] X. Li and G. Chen, “Synchronization and desynchronization of
complex dynamical networks: an engineering viewpoint,” IEEE Trans.
variation of the second-largest eigenvalue of the coupling
On Circ. and Sys. I, vol. 50, no. 11, Nov. 2003, pp. 1381-1390.
matrix for Fig. 4(a) and (b) with the time, respectively. Fig. 5 [8] S. F. Bush and A. B. Kulkarni, “Genetically induced communication
shows that the whole network spends six steps for network fault tolerance,” SFI Workshop: Resilient and Adaptive
self-healing, and the second-largest eigenvalue of the Defence of Computing Networks, vol. 9, no. 2, 2003, pp. 19-33.
[9] T. Arai, E. Pagello and L. E. Parker, “Editorial: Advances in
coupling matrix decreases, which indicated that the proposed multi-robot systems,” IEEE Trans. on Robotics and Automatic, vol. 18,
topology control for self-healing is effective. no. 5, Oct. 2002, pp.655-661.
[10] J. A. Fax and R. M. Murray, “Information flow and cooperative control
of vehicle formations,” IEEE Trans. on Automatic Control, vol. 49, no.
VI. CONCLUSIONS 9, Sep. 2004, pp. 1465-1476.
Self-healing in mobile robot networks refers to the process [11] W. Sheng, Q. Yang, S. Zhu and Q. Wang, “Efficient map
synchronization in ad hoc mobile robot networks for environment
of maintaining not only logical topology but also physical one exploration,” In Proc. of IEEE/RSJ Conf. on Intelligent Robots and
of the network. We first establish an interaction dynamics Systems (IROS), Edmonton, Canada, Aug. 2005, pp. 1524-1529.
model describing both logical and physical topologies of the [12] V. Gazi and K. M. Passino, “Stability analysis of swarms”, IEEE Trans.
on Automatic control, vol. 48, no. 4, April 2003, pp. 692-697.
network. Considering the mobility of mobile robot network,
[13] R. Olfati-Saber and R. M. Murray, “Consensus problems in networks of
we provide a recursive, distributed topology control for agents with switching topology and time-delays,” IEEE Trans. On Auto.
self-healing when mobile robots fail, and give a metric of the Control, vol. 49, no. 9, Sept. 2004, pp.1520-1533.
topology structure for evaluating the performance of [14] D. S. Callway, M. E. J. Newman, S. H. Strogatz, and D. J. Watts,
“Network robustness and fragility: percolation on random graphs,”
recovered network topologies. Based on Lyapunov exponent, Phys. Rev. Lett., vol. 85, no. 25, 2000, pp. 5468-5471.
we prove the global stability of the network with the fully [15] J. M. Carlson, and J. Doyle, “Complexity and robustness,” Proc. of the
distributed topology control. Finally, the results of National Academy of Sciences of USA, Suppl. 1, 1999, pp. 2538-2545.
[16] M. Barahona, and L. M. Pecora, “Synchronization in small-world
simulations have demonstrated the validity of the proposed systems,” Phys. Rev. Lett., vol. 89, no. 5, 2002, pp. 054101.
modeling and control methods. [17] R. Diestal, Graph Theory, New York, NY: Springer-Verlag, 1997.
The conclusions about self-healing in mobile robot [18] F. Xue and P. R. Kumar, “The number of neighbors needed for
connectivity of wireless networks,” Wireless Networks, vol. 10, no. 2,
networks can be also applied in mobile sensor networks. March 2004, pp. 169-181.
[19] X. F. Wang and G. Chen, “Synchronization in scale-free dynamical
REFERENCES networks: Robustness and fragility,” IEEE Trans. on Circ. and Syst.-I,
vol. 49, no. 1, 2002, pp. 54-62.
[1] F. Zhang, W. Chen and Y. Xi, “Motion synchronization of mobile robot
[20] P. Santi, Topology Control in Wireless Ad Hoc and Sensor Networks,
networks: robustness,” in Proc. of IEEE/RSJ Conf. on Intelligent
John Wiley & Sons, Sep. 2005.
Robots and Systems (IROS), Beijing, China, Oct. 2006, pp. 5570-5575.
[21] G. Rangarajan and M. Z. Ding, “Stability of synchronized chaos in
[2] E. Shaw, “The schooling of fishes,” Scientific American, vol. 206, 1962,
coupled dynamical systems,” Phys. Lett. A, vol. 296, 2002, pp. 204-209.
pp. 128–138.
[3] C. W. Reynolds, “Flocks, herds, and schools: A distributed behavioral
model,” Computer Graphics (ACM SIGGRAPH ’87 Conference
Proceedings), vol. 21, no. 4, 1987, pp. 25-34.
[4] T. Vicsek, A. Cziroók, E. Ben-Jacob, I. Cohen, and O. Shochet, “Novel
type of phase transition in a system of self-driven particles,” Physical
Review Letters, vol. 75, no.6, 1995, pp.1226-1229.
[5] J. K. Parrish, S. V. Viscido, and D. Grunbaum, “Self-organized fish
schools: an examination of emergent properties,” Boil. Bull., vol. 202,
2002, pp. 296-305.

3112