A Distributed Self Spreading Algorithm

for Mobile Wireless Sensor Networks

Nojeong Heo and Pramod K. Varshney
Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY 13244,USA
{nheo,varshney}@syr.edu

Abstract - Sensor deployment is an important problem in

mobile wireless sensor networks. This paper presents a
distributed self deployment algorithm for mobile sensors.
Performance metrics to evaluate algorithm performance are
coverage, uniformity, time and distance traveled till the
algorithm converges. Our algorithm is compared with a
simulated annealing based algorithm for deployment and is
shown to exhibit excellent performance.

1. Introduction
Design and deployment of infrastructured networks such as a
cellular network has matured over the last two decades. In such
networks, mobile users access the network via fixed base stations.
Planning and deployment of these networks is carried out based on
radio propagation and terrain models with the goal of maximizing
radio coverage. More recently, there has been a great deal of
interest in ad hoc wireless networks. These networks employ fixed
or mobile nodes and dynamically organize themselves into a
network without requiring an infrastructure. In ad hoc networks,
each node acts not only as an end node but also as a router. One
important aspect in the design of these networks is the initialization
procedure and establishment of routing structure. Most research on
ad hoc networks has been focused on issues such as the
development of routing protocols and quality of service and not on
topology and deployment.
Wireless sensor networking has become an area of intense
research activity. This is due to technical advances in sensors,
wireless communications and networking, and signal processing.
Many applications are envisaged including environment
monitoring, battlefield surveillance, and urban search and rescue
especially in hazardous situations. Wireless sensor networks
(WSN) operate under limited radio coverage and attempt to
conserve bandwidth and battery power. Most of the research on
WSN has focused on the development of collaborative signal
processing and power aware algorithms [1]. Sensor nodes are
generally assumed to be fixed and randomly placed because of
practical reasons. The number of sensors is assumed to be quite
large so that coverage of the surveillance area is not an issue. Not
much attention has been paid to optimization in terms of number of
nodes or their topology.
Recently, Meguerdichian et al. have considered the problem of
location and deployment of sensors in a WSN from a coverage
standpoint [2]. They implicitly assumed fixed wireless sensor
nodes. They argued that coverage is a primary performance metric
that provides an indication regarding quality-of-service. They

combined computational geometry and graph theoretic approaches to

develop algorithms for coverage calculations. We also focus on
coverage in WSN and discuss it later in the paper. Bulusu et al.s work
[3] is somewhat similar to the deployment problem that is considered
here. They have investigated the problem of adaptive beacon placement
for localization in a WSN. They also pointed out the lack of viability
and inadequacy of fixed and dense beacon placement in some situations
due to node perturbation during deployment, noisy environment, and
self-interference. By placing additional beacons incrementally, they
achieve empirical adaptation to terrain conditions.
Another related problem is the Art Gallery Problem in computational
geometry [4]. The art gallery problem tries to find the minimal number
of positions for guards or cameras so that every point in a gallery is
observed by at least one guard or camera. A deterministic solution can
be found for the art gallery problem and it appears to be a possible
solution to a variety of sensor placement problems. Even though there
are many solutions to the art gallery problem, all of them assume the
availability of a good model of the environment a priori. However, it is
virtually impossible to have complete information of the environment in
a WSN. Too much communication over long range to obtain global
information requires a huge amount of energy. This is an unaffordable
burden on a system with limited power supply. Thus, deterministic
deployment is impractical due to many reasons such as the harshness of
deployment region that may be remote and inhospitable and the
increased cost and latency due to the huge number of nodes deployed
[5].
In our work, we are interested in the self-deployment of mobile
sensor nodes. This is quite similar to problems considered in
cooperative mobile robotics [6]. Mobile sensors are often desirable
since they can patrol a wide area, and they can be re-positioned for
better surveillance [7]. Some researchers considered the use of mobile
robots in sensor networks. Winfield [8] considered autonomous
dispersion of mobile nodes in a scenario where mobility is required to
cover the entire region due to a lack of wireless network connectivity.
He used a random diffusion method for node deployment while
collecting data over a fixed surveillance region. In the incremental
deployment algorithm [9], nodes are added one at a time. The goal is to
maximize network coverage under the constraint that nodes maintain
line-of-sight with each other. Loo et al. considered a system consisting
of a number of cooperating mobile nodes that move toward a set of
prioritized destinations under sensing and communication constraints
[10]. They show how individual agents know when cooperation
between agents improves the performance and when they should
suspend cooperation.

Much research has been done about various issues related to the
deployment problem. Most approaches use either a centralized
solution as in the circle covering problem [11,12] and the
geometric problem [13], or are restricted to a certain topology as in
the coding theoretic approach [14]. Since some approaches adopt
random deployment as in the set covering problem [5] and the
topology discovery problem [15], initial distribution determines the
utilization of networks.
Our work is different from prior work on the deployment
problem. Our deployment algorithms main objective is topology
improvement for longer system lifetime by utilizing mobility of
robots. We provide a decision and control mechanism at each robot
to be used during deployment rather than random diffusion, which
is used in Winfields work. In contrast to Howard et al. who use an
incremental approach, the nodes in our algorithm are deployed at
the same time and they organize themselves in an adaptive manner.
Unlike Loo et al., our algorithm does not require prespecified
destinations to form an energy efficient topology.
Our self deployment algorithm will be more useful in situations
where it is hard to ensure precise initial deployment due to the fact
that the deployment area is too dangerous or inaccessible to
humans. Randomly scattered sensors over a battlefield or a
hazardous site are not likely to form a uniform distribution and
provide desired coverage. Modification of WSN topology in an
autonomous and distributed manner using our algorithm can help
in improving sensor coverage and also to prolong expected system
lifetime. This is essential in time-critical applications. For example,
if some area is contaminated by a hazardous material, a properly
deployed sensor network can quickly sense and measure the
amount of hazardous material such as poisonous gas or nuclear
leakage. By fully covering the entire area of interest, the overall
condition can be assessed quickly and this information can be used
for search and rescue missions as well as for evacuation route
planning.
The main goals of this paper are
1) discussion of the issue of coverage in WSN in detail
2) development of a distributed algorithm for self-deployment
In the next section, we discuss performance metrics for a mobile
WSN. In Section 3, we formulate the sensor deployment problem.
Our algorithm is presented in Section 4 followed by simulation
results in Section 5. Some concluding remarks are provided in
Section 6.

2. Performance Metrics in Mobile WSN

Selection of suitable measures to compare performances of
different approaches is an important issue in a mobile WSN.
Coverage, uniformity, time, and distance are considered as
performance metrics in mobile wireless sensor networks here.
Coverage and uniformity are related to the performance of sensor
networks after the deployment of sensors is complete. Time and
distance are directly related to the performance of the deployment
scheme itself.
Coverage
Generally, coverage can be considered as the measure of quality
of service of a sensor network. The concept of coverage as a
paradigm for the system level functionality of multi-robot systems
was introduced by Gage [16]. Gage described three types of
coverage behavior for problems such as detection of targets over a
surveillance area. These behaviors are: blanket coverage, where the
objective is to obtain the maximum detection rate over a given

area; barrier coverage, where the objective is to minimize the

probability of miss through the barrier; and sweep coverage, which is
roughly equivalent to a moving barrier. Our self deployment algorithm
can be classified as a blanket coverage problem according to Gages
taxonomy. Coverage can be classified as either deterministic or
stochastic by the manner in which nodes are deployed. To obtain
deterministic coverage, sensor nodes should be placed at predefined
locations. The art gallery problem is an example of deterministic
coverage. By dispersing sensors randomly in the environment,
stochastic coverage can be achieved.
In this paper, coverage is defined by the ratio of the union of covered
areas of each node and the complete area of interest. Here the covered
area of each node is defined as the area within sensing radius R. Perfect
detection of all interesting events in the covered area is assumed.
C=

U Ai

i =1,..., N

where

Ai is the area covered by the ith node,

N is the total number of nodes,
A stands for the area of the region of interest(ROI).

If a node is located well inside the region of interest, the complete

circle will lie within the ROI. In this case, the full area of that circle,
i.e., R 2 , is counted as the covered region. If a node is located near the
boundary of the ROI, then only the partial area of the ROI covered by
that node is included in the computation. Because of areas covered by
nodes that fall out of the ROI and the fact that overlap of covered areas
between nodes should not be included while computing coverage, we
need to use more nodes than simply the ratio of A and area sensed by a
single node. We distinguish sensing range and communication range of
a node. In general, they will be different and accordingly sensing
coverage and communication coverage will be different. Sensing
coverage can be accrued when sensor nodes are connected via wireless
links.
Uniformity
Uniformly distributed sensor nodes spend energy more evenly
through the WSN than an irregular topology does. So, uniformity of
network topology can be a good estimator for the expected system
lifetime. Also, fewer nodes are required to cover an ROI when nodes
are more evenly distributed. Uniformity can be defined by the average
local standard deviation of the distances between nodes.

U=

1
N

Ui = (
where

i =1

1
Ki

Ki

( Di, j M i )2 ) 2
j =1

N is the total number of nodes

Ki is the number of neighbors of the ith node,
Di,j is the distance between ith and jth nodes,
Mi is the mean of internodal distances between the
ith node and its neighbors.

In the calculation of the local uniformity Ui at the ith node, only

neighboring nodes that reside within its communication range are
considered. The uniformity measure is a local measure and is computed
locally because each node has access to local information only. A
smaller value of U means that nodes are more uniformly distributed in
the ROI. In uniformly distributed networks, internodal distances are
almost the same; the expected energy consumption per communication
as well as the expected lifetime of each node is almost the same.
Therefore, we can expect full energy utilization at each node and longer
system lifetime for uniformly distributed networks.

Time
The time spent for deployment is also important in many timecritical applications such as search and rescue and disaster
recovery operations. Mostly, the required time depends on the
complexity of the reasoning algorithm and physical time for the
movement of nodes. The total time elapsed is defined here as the
time elapsed until all the nodes reach their final locations. We
focus here on the time spent for deployment itself and not on data
transmission delays from a source node to a destination node that
is commonly used for network performance evaluation and its
quality of service.
Distance
The average distance traveled by each node is related to the
required energy for its movement. So, the expected distance
traveled is important for the estimation of required energy (fuel)
when each node has a limited energy supply. The variance of
traveled distance is also important to determine the fairness of the
deployment algorithm and for system energy utilization.

without any intervention from a central controller that acts remotely.

We call this algorithm Distributed Self-Spreading Algorithm(DSSA)
and discuss it in detail.
To begin with, a specified number of nodes are pre-deployed
randomly in a given region, for instance, inside a rectangle. The sensing
range (sR) and the communication range (cR) are assumed to be given.
Each node can sense or detect an event within its sensing range and any
pair of nodes within their communication range can communicate with
each other. This communication is needed for finding neighborhood,
obtaining locations of nodes in the neighborhood, and transmitting and
forwarding sensed data. Neighborhood is defined here as nodes within
the communication range. The flow chart of the algorithm is given in
Figure 1. This distributed algorithm is executed at each node i.
Start
Initialization
p0, sR, cR
Calculation of density
,D0, D=D0

3. Mobile Node Placement/Deployment Problem

Calculation of partial force

fni,j(,D,cR,pn)

3.1. Assumptions and restrictions

We assume that all sensor nodes have capabilities for sensing,

communication, computation, and mobility. Sensing coverage and
communication coverage of each node is assumed to be ideal,
which means that both coverage areas have a circular shape
without any irregularity. Computation capability is required at each
node to support a distributed algorithm that includes a reasoning
process for deployment and routing. We assume that the initial
deployment is random and a distributed deployment algorithm will
be executed starting from the initial random topology using each
nodes mobility. Another assumption is that every node has the
ability to know its own location by some method such as GPS or
iterative multilateration [17]. This locationing ability is needed by
each node while making a decision regarding its next movement in
the deployment process. Also, we assume that there are no errors
during transmissions of data and in the calculation of locations.

Temporary position
pin+1' = pin + sum(fni,j)

Oscillation?
|pin-1-pin+1'| < e

Covering and placement problem can occur in many

applications from environment monitoring to battlefield
surveillance. Without loss of generality, we can consider the
covering problem in a rectangular ROI with a certain number of
nodes that form an ad hoc wireless network. The goal is to find
positions and movements of nodes to achieve maximum coverage
and to form a uniformly distributed wireless network in minimum
time and with minimum energy consumption. We develop a
heuristic algorithm and evaluate its performance in terms of these
performance metrics.

4. Distributed Self-Spreading Algorithm

The algorithm is inspired by the equilibrium of molecules,
which minimizes molecular electronic energy and inter-nuclear
repulsion. Each particle obtains its own lowest energy point in a
distributed manner and its spacing from the other particles is
almost the same. While deploying a wireless sensor network using
mobile nodes, we observe that a similar phenomenon is needed.
We propose an algorithm that can cover the region of interest

Count oscillation
Ocount = Ocount+1

No

pin+1=pin+1'
Update D

Yes

Ocount < Olim

No

No

Stable?
|pin+1-pin| < e

pin+1=1/2(pin+pin+1')
Update D

Yes

Count stable status

Scount = Scount+1

The restriction is that each node has only local information from
the neighboring nodes within its direct communication range. The
communication range of each node is defined by the maximum
distance at which the signal to noise ratio is above the threshold
required for achieving the design goal in terms of power
conservation.

3.2. Problem formulation

Yes

Scount < Slim

Yes
No

pin+1 = pin

Stop

Fig. 1 Distributed Self-Spreading Algorithm (DSSA)

The algorithm begins with the specification of cR, sR, and the initial
node locations (p0). In our algorithm, we require the quantity called
expected
density.
This
can
be
calculated
by
using

(cR) =

N cR 2 , where N is the number of nodes and cR is the

A

communication range of each node, and A is the ROI. Thus, expected

density is the average number of nodes required to cover the entire area
when these nodes are deployed uniformly. Initial local density D0 of a
node is equal to the number of nodes within its communication range.
These densities will be used when decisions regarding positions of
nodes are made.
We introduce the concept of force to define the movement of nodes
during the deployment process. The force is dependent on the distance
between nodes and the current local density. The force from a node that
is closer is greater than that from a node that is farther just like the
particles in Physics that follow Coulombs law.

We define a force function that satisfies the following conditions.

(i) Inverse relation: f(d1) f(d2), when d1 d2 , where d1 and
d2 are node separations from the origin.
(ii) Upper limit: f(0+) = fmax.
(iii) Lower limit: f(d) = 0, where d > dth, d is the node
separation and dth is the threshold to define the local
neighborhood.
Condition (i) is the same as in Physics, but conditions (ii) and (iii)
are included to modify the model to incorporate the notion of
locality. In other words, a limiting function is applied via
conditions (ii) and (iii).
The partial force at time step n on the ith node from the jth node
that is in the neighborhood of the ith node is calculated in this
paper as

f ni , j =
where

(cR | p ni p nj |)

p nj p ni
| p nj p ni |

(1)

cR stands for communication range

p ni

stands for the location of ith node at time step n

The density factor, which is defined as the ratio of the local

density (D) and the expected density () at each node, is small in
sparse regions and is large in dense regions. Also internodal
distance affects the partial force inversely. Closely located nodes
impose larger partial forces and nodes that are far apart induce
smaller partial forces.
After adding all the partial forces at the current location, each
node can decide its next movement. The local information is
collected from the nodes that are within the communication range
and that information is used for the calculation of the local density
at each node. Each nodes movement is decided by the combined
force at that node due to nodes in its neighborhood.
When should a node stop its movement is an important issue.
Two stopping criteria are introduced in the Distributed SelfSpreading Algorithm. If a node moves less than e for the time
duration Slim, this node can be considered to be in the stable status
and that node stops its movement. This stopping criterion is for
stationary nodes because of either empty fuel or broken mobile
units and also for the nodes that have reached the stable status. If a
node moves back and forth between almost the same locations
many times, this node is regarded in the oscillation status. By
comparing with the history of its movement, each node can know
if oscillation is going on. Then one counts the number of
oscillations and if this oscillation count (Ocount) is over the
oscillation limit (Olim), we stop that nodes movement at the center
of gravity of the oscillating points.
When sensor nodes are deployed in a remote and hostile region,
some nodes can be affected during and after deployment period.
Some nodes can lose their mobility and other nodes can lose their
communication functionality. So we need a robust deployment
method that can handle these difficulties. Our algorithm exhibits
this kind of robustness. First, when the sensor node loses its
mobility, that sensor node does not move and is considered to be
an early stopped node. Neighboring nodes, if they can move, may
still improve the irregular topology. Second, when a sensor node
loses its communication capability, that sensor node is of no use in
a sensor network. Because each nodes movement is only affected
by the current status of neighboring nodes, each node is adaptive to
environment changes such as node failures, various terrain shapes,
etc.

5. Simulation Results
We consider 30 randomly placed nodes in a region of size 10 10 to
run the Distributed Self-Spreading Algorithm. We assume sR=2 and
cR=4.
Fig.2 shows the locations and coverage of the initial deployment
before running the Distributed Self-Spreading Algorithm. Tiny circles
represent the positions of nodes and small (shaded) and large circles are
used to show the sensing range and the communication range
respectively. Sensor information may be collected within the sensing
range and communications between nodes are possible within the
communication range. Communications are possible between nodes that
are connected by a line in the figure. As seen in Fig 2., some parts of the
region cannot be covered by the nodes that are randomly dispersed,
even though there are enough numbers of them in the given ROI. In this
particular example, the network is not fully connected, so the actual
coverage is much smaller than just adding the entire covered region.
The calculated coverage is more than 90%, but the actual coverage is
well below 50% because the network is separated. This situation is
exactly the case where topology improvement is required.
Iintial position of 30 sensors, C = 0.93483
12

10

-2
-2

10

12

Fig. 2 Initial distribution of sensor nodes

Fig. 3 shows nodes location and coverage after running the
Distributed Self-Spreading Algorithm. The rectangle is fully covered
after running the algorithm. The parameter values used in this
simulation run are: stable status limit(Slim) =10, oscillation limit (Olim) =
10, and threshold (e) for oscillation and stable status =0.25994. Now the
network is fully connected and also can cover the entire ROI. Note the
spatial node distribution is more uniform than the initial random
distribution in Fig. 2.
Iteration # : 36, Covered Area: 1, Slim: 10, Olim: 10, e: 0.25994
12

10

-2
-2

10

12

Fig. 3 Final distribution after running DSSA

For the purpose of comparison, Simulated Annealing was also

used for self-spreading. Simulated Annealing algorithm is known
as a good solution of many combinatorial optimization problems.
To implement a Simulated Annealing algorithm, 4 main design
issues should be considered. These are: the definition of the
neighborhood, move operator, local energy calculation, and
annealing schedule. The definition of neighborhood is the same as
for DSSA, i.e., it is set equal to the communication range. This
concept of neighborhood is reasonable because each node can only
reach the neighboring node in a single hop communication. The
move operator is chosen to be a random movement within the
neighborhood. The local energy calculation is done by adding up
sub-forces in the neighborhood just like our algorithm. The
exponential cooling schedule is used as the annealing schedule for
efficiency as in [18]. The parameters used are initial temperature
T0=1 and as the stopping criterion 3 consecutive failure of
achieving the desired acceptance ratio is used. The result after
applying the simulated annealing algorithm is shown in Fig. 4. Fig.
4 shows that Simulated Annealing also works well for this initial
distribution. The entire area is covered by 30 sensor nodes and
these nodes are well spread over the region.
Iteration # : 19, Covered Area: 1, T0: 1, failureIndexLimit: 3
12

10

InitstdSeparation, stdSeparation
1.6
1.4
1.2
1
0.8
0.6
0.4

Initial
SA

0.2
0
10

Our
15

20

25

30
# of Nodes

35

40

45

50

Fig. 6 Uniformity versus network size

Fig. 5 shows the improvement in coverage area from the initial
random deployment for both algorithms; simulated annealing and ours.
Though both algorithms have quite similar performance, our algorithm
has better coverage for smaller number of nodes. As the number of
nodes increases, the improvement in coverage diminishes. Standard
deviation of internodal distances can be considered as the metric of
uniformity of the networks. Fig. 6 shows the reduction in the standard
deviation from the initial case. Both algorithms obtain better uniformity
than the initial one and ours outperforms simulated annealing. The
improvement in uniformity is not that different with varying network
density.

stoppedNum
35

30
0

25
-2
-2

10

12

20

Fig. 4 Final distribution after running SA

SA

15

The performance of our algorithm is evaluated in terms of the

metrics presented in Section 2. Results are presented in Fig. 5 ~ 8.
These results are obtained for different number of nodes dispersed
over a fixed ROI of size 10 10, i.e., for different node densities.
Number of nodes varies from 10 to 50 and results are averaged
over 100 runs.

Our
10

0
10

15

20

25

30
# of Nodes

35

40

45

50

Fig. 7 Termination times versus network size

InitArea & Covered Area
1
meanTotalDistance

Our
0.95

80

SA
70

0.9

60
50

Initial Area

0.85

SA

40
0.8
30
0.75

20
10

0.7
10

15

20

25

30
# of Nodes

35

40

45

Fig. 5 Coverage versus network size

Our

50
0
10

15

20

25

30
# of Nodes

35

40

45

50

Fig. 8 Distance traveled versus network size

Fig. 7 shows that our algorithm leads to faster deployment than

simulated annealing. Also termination times of our algorithm are
similar over a wide range of number of nodes. This means that our
algorithm is more insensitive to the number of nodes, i.e., network
density.
Fig. 8 shows mean distance traveled to reach the final locations
for deployment. Our algorithm requires less distance traveled than
the simulated annealing algorithm. This distance is related to the
required energy (fuel) for deployment. Also, we observe that the
required energy as well as the distance traveled at different node
densities is almost constant. Thus, the required energy (fuel) is
quite insensitive to network density.
As seen in Fig.5~Fig.8, 25~40 nodes are required to attain
acceptable performance. When too few nodes are used, we cannot
obtain full coverage over the region of interest. When too many
nodes are used, we do not gain that much coverage improvement
because of the diminishing marginal gain in terms of coverage,
though we can still obtain more uniform distribution. With the
number of nodes in this range, the required time and distance of
our algorithm is much smaller than that of the Simulated
Annealing algorithm. Because the variation of required time and
traveled distance is small over this range of node densities, it is
easy to estimate the required energy for deployment.

6. Conclusions
We considered the sensor coverage problem for the deployment
of wireless sensor networks here. A region of interest needs to be
covered by a given number of nodes with limited sensing and
communication range. We start with a random distribution of the
nodes over the region of interest. Though many scenarios adopt
random deployment because of practical reasons such as
deployment cost and time, random deployment may not provide a
uniform distribution which is desirable for a longer system lifetime
over the region of interest. In this paper, we propose a distributed
algorithm for the deployment of mobile nodes to improve an
irregular initial deployment of nodes. After going through the
algorithm, the area of interest is covered by uniformly distributed
nodes. While developing the algorithm, one should consider
factors such as density of nodes, memory constraints, localization
errors, and scalability of mobile nodes. Through the requirement of
mobility and locationing ability of nodes, this algorithm provides a
way to avoid expensive redeployment process. This postdeployment idea will be more useful especially when a large
fraction of nodes are destroyed or broken during deployment or in
a hostile situation, where initial distribution is quite uneven and
when redeployment is too costly or too risky. The performance of
the algorithm is determined by the percentage of region covered,
by computational/deployment time, by the mean distance that is
required for the deployment, and by the uniformity of the networks.
Simulation results show that our algorithm successfully obtains a
uniform distribution from initial uneven distribution. The
performance of our algorithm is compared with the Simulated
Annealing algorithm and exhibits excellent performance.

Acknowledgments
This work was supported in part by the National Science
Foundation under grant number ECS-9901361 and by NASA
under grant number NAG5-11227.

References
[1] Kumar, S.; Feng Zhao; Shepherd, D., Collaborative signal and
information processing in microsensor networks, IEEE Signal
Processing Magazine, vol.19, no. 2, pp. 13 -14, Mar. 2002.

[2] Seapahn Meguerdichian, Farinaz Koushanfar, Miodrag Potkonjak,

and Mani Srivastava, Coverage Problems in Wireless Ad-hoc Sensor
Networks, IEEE Infocom, pp. 1380-1387, 2001.
[3] Nirupama Bulusu, John Heidemann, and Deborah Estrin, Adaptive
Beacon Placement, in Proceedings of the 21st International
Conference on Distributed Computing Systems, pp. 489-498, Phoenix,
AZ, Apr. 2001.
[4] O'Rourke, J., Art Gallery Theorem and Algorithms. New York:
Oxford University Press, 1987.
[5] S. Slijepcevic and M. Potkonjak, Power efficient organization of
wireless sensor networks, Proceedings of IEEE International
Conference on Communications, 2001.
[6] Cao, Y. U., Fukunaga, A., and Kahng, A., Cooperative mobile
robotics: Antecedents and directions, Autonomous Robots, vol. 4, pp.
123, 1997.
[7] H. Qi, S. S. Iyengar and K. Chakrabarty, Distributed sensor fusiona review of recent research, Journal of the Franklin Institute, vol. 338,
pp. 655-668, 2001.
[8] A.F.T. Winfield, Distributed sensing and data collection via broken
ad hoc wireless connected networks of mobile robots, in Distributed
Autonomous Robotic Systems 4, eds. LE Parker, G Bekey & J Barhen,
Springer-Verlag, pp. 273-282, 2000.
[9] Andrew Howard, Maja J Mataric and Gaurav S. Sukhatme, An
Incremental Self-Deployment Algorithm for Mobile Sensor Networks,
To appear in Autonomous Robots Special Issue on Intelligent Embedded
Systems, 2002.
[10] Lit-Hsin Loo, Erwei Lin, Moshe Kam and Pramod Varshney,
Cooperative Multi-Agent Constellation Formation under Sensing and
Communication Constraints, in Cooperative Control and Optimization,
pp.143-170, Kluwer Academic Press, 2002.
[11] J. B. M. Melissen, P. C. Schuur, Improved coverings of a square
with six and eight equal circles, Electronic Journal of Combinatorics,
vol.3, no.1, 1996.
[12] K. J. Nurmela, P. R. J. stergrd, Covering a square with up to 30
equal circles, Research Report A62, Laboratory for Theoretical
Computer Science, Helsinki University of Technology, 2000.
[13] K. Sugihara and I. Suzuki, Distributed Algorithms for Formation
of Geometric Patterns with Many Mobile Robots, Journal of Robotic
Systems, vol.13, no.3, pp. 127-139, 1996.
[14] K.Chakrabarty, S.S.Iyengar, H.Qi, E.Cho, Coding theory
framework for target location in distributed sensor networks, IEEE
International Conference on Information Technology: Coding and
Computing, June 2001.
[15] Roger Wattenhofer, Li Li, Paramvir Bahl, Yi-Min Wang,
Distributed Topology Control for Wireless Multihop Ad-hoc
Networks, Proceedings IEEE INFOCOM 2001, pp. 1388-1397, 2001.
[16] Gage, D. W., Command Control for Many-Robot Systems,
AUVS-92, the Nineteenth Annual AUVS Technical Symposium,
Huntsville AL,22-24 June 1992. Reprinted in Unmanned Systems
Magazine, Vol. 10, No. 4, pp 28-34, Fall 1992.
[17] K. Sohrabi, B. Manriquez, and G. Pottie, Near-ground wideband
channel measurements, In Proceedings of the 49th Vehicular
Technology Conference (Houston, May 1620), IEEE, New York,
1999, pp. 571574.
[18] S. Kirkpatrick, C.D. Gelatt, and M.P. Vecchi, Optimization by
simulated annealing, Science, vol. 220, pp. 671-680, 1983.