Wireless Sensor Network, 2010, 2, 267-273
doi:10.4236/wsn.2010.24037 Published Online April 2010 (http://www.SciRP.org/journal/wsn)
A Sensing and Robot Navigation of
Hybrid Sensor Network
Shuncai Yao1, Jindong Tan2, Hongxia Pan1
1Department of Electric Engineering, North University of China, Taiyuan, China
2Department of Electrical Computer Engineering, Michigan Technological University, Houghton, USA
Received June 9, 2009; revised June 25, 2009; accepted June 29, 2009
Abstract
Traditional sensor network and robot navigation are based on the map of detecting fields available in ad-
vance. The optimal algorithms are explored to solve the energy saving, shortest path problems, etc. However,
in practical environment, there are many fields, whose map is difficult to get, and need to detect. This paper
explores a kind of ad-hoc navigation algorithm based on the hybrid sensor network without the prior map.
The system of navigation is composed of static nodes and mobile nodes. The static nodes monitor events
occurring and broadcast. In the system, a kind of cluster broadcast method is adopted to determine the robot
localization. The mobile nodes detect the adversary or dangerous fields and broadcast warning message.
Robot gets the message and follows ad-hoc routine to arrive the events occurring place. In the whole process,
energy saving has taken into account. The algorithms of nodes and robot are given in this paper. The simu-
late and practical results are available as well.
Keywords: Robot Navigation, Hybrid Sensor Network, Routine Search
1. Introduction
Sensor network is an advance technology in recent years.
There have been many papers in this field. Most of these
are concerned in static sensor network or mobile sensor
network separately. The communication and related
technology have been studied. The static sensor network
consists of the fixed nodes and adopts the ad-hoc com-
munication. This kind of system can monitor the envi-
ronment and send the messages to the base station. En-
ergy would be saved, but the system can hardly do any-
thing better if the whole condition is changed or the ad-
versary available. In mobile sensor network, the nodes
can adjust their places according the algorithms, which
are based on the changing condition. However, at the
same time, the energy would be cost much. It is well
know that is a serious problem for sensor networks.
In navigation, most of these two systems always adopt a
method that is based on mapping. That is to say, the whole
environment map is known in advance. In spite that the
optimal algorithms have been adopted in these systems, it
is not so suitable for the genuine conditions yet.
In this paper, we explore a hybrid sensor network. It is
composed of both static sensor nodes and mobile sensor
nodes. The static sensor nodes are deployed regularly,
and detect the events of environment. While, the mobile
sensor nodes detect the adversary or dangerous field.
When these fields are found the mobile nodes will stop at
the edge of adversary or dangerous fields and send a
message to robot. Finally, there will be more nodes at the
edge of adversary or dangerous fields to protect the robot
not to approach. The method is a compromise between
the energy saving and flexible of whole system. Based
on this scheme, non-mapping navigation comes into real-
ity. It is proved that it is a better scheme in energy saving
and flexibility through the simulate experiment at the end
of this paper.
2. Related Work
In a static sensor network, the quantity of nodes is an
important problem. The more nodes means the more cov-
erage of whole area and more accurate of navigation.
Lots of research has focused on the sensor localization
[1], structural monitoring [2] or protocols [3]. Recently,
researchers have begun to explore the method, which low
down the quantity. And, it is taken into account that the
adversary and dangerous fields available in environment.

S. C. YAO ET AL. 268
Since these methods adopt the skeleton graphs and opti-
mal algorithm, the quantity of nodes and the price of
communication have been lowed down.
The quantity of nodes is not such a pivotal problem in
mobile sensor network because its flexibility. The problem
of energy costs is tried to low down by sorts of algorithm
based on the passive algorithm [4,5] or statistic-learning
algorithm. On the other hand, these algorithms are com-
plicated. They are not so easy to come into realities and
the use more capacity of nodes memory.
The sensor network of this paper develops the hybrid
sensor network and adopts a kind of cluster navigation.
At first, the static sensor nodes are deployed as regular as
possible in the whole environment. That is to say, the
localization of static sensor nodes should be arranged at
the correct place, but we need not care it so much. These
static sensor nodes may local in the adversary or dan-
gerous fields, but it does not matter. We can take meas-
ure to solve the problem. The sensing program can be
downloaded in these sensor nodes. They can sense the
change of environment area that we care. The mobile
sensor nodes, actually they are the small mobile robots,
but they just do the dangerous fields detecting work and
nothing else. Navigated robots are not these nodes. Actu-
ally, there is another robot being navigated. In Figure 1,
the static sensor nodes are deployed in the whole area.
They have sensing and communication abilities. Actually,
these nodes are tmote Sky [6] made by Moteiv Corpora-
tion. They are ultra low power wireless module for use in
sensor networks, monitoring applications, and rapid ap-
plication prototyping. Tmote Sky leverages industry
standards like USB and IEEE 802.15.4 to interoperate
seamlessly with other devices.
Figure 1. The structure of system.
The mobile sensor nodes can gather around the edge
of adversary or dangerous. The robot is in a waiting
status. The mobile nodes are four-wheel cart, which
carry the Mica nodes. The Mica wireless platform
serves as a foundation for the emerging possibilities.
With sensing, communication, and I/O capabilities, Mica
can simultaneously act as a data router, sensor interface,
and control point. Mica’s flexible design serves as a
building block for creating efficient application-specific
protocols. And Mica does not require use of predefined
protocols [7].
We have done many works both on the communica-
tion of static sensors and moving of mobile sensor nodes.
This is the base that we can explore the scheme.
3. The Deployment of Static Sensor Nodes
and Communication Algorithm
The static sensor nodes deploy all the detecting area. In
view of the energy saving, we adopt a kind of street map
and skeleton graph [8]. This scheme constructs a reduced
graph with as few as possible working nodes instead of
the whole area covering static sensor nodes. All the
nodes stand at the key points of the detecting area. They
are the working nodes. That does not mean we do not
need large number of static nodes. If the quantity of
static nodes is not enough, the navigation can become
very difficult, we discuss this problem at navigation. The
nodes, which are not on the key points, fall in sleep. That
means these sleeping nodes are working at the low power
state. When the system needs them, they will be waken
up. These sleeping nodes are in working state, and afford
robot navigation. That is to say, there is large number of
nodes in system, but, the working nodes are not so many
and energy is saved sequentially. On the other hand, the
flexibility and accuracy of system can be maintained.
Those nodes, which are not in adversary fields and not
on the point, sleep unless the robot navigation needs
them (We will discuss in Section 5.).
Except these non-key point nodes, there are still some
nodes falls into sleep. They are the nodes lie in the ad-
versary or dangerous fields. They can keep the sleeping
status through the all process. Actually, they have quitted
events monitor. Therefore, the street map and skeleton
graph can be described as Figure 2.
Given such preliminary works, we modify the static
nodes algorithm to arrange the nodes places, monitor events
and broadcast the message as shown in Algorithm 1.
Algorithm 1. Deploy of static sensor nodes
1: while TRUE do
2: detect the location
3: if location ! = dangerous fields & location ! = key
point then
4: send navigation requirement
5: else
Mobile sensor nodes Static sensor nodes
N
avigated robo
t

S. C. YAO ET AL.269
Figure 2. The diagram of on point nodes and the sleeping
nodes. These nodes which is in the dangerous fields are put
into sleeping status.
6: node keeps sleep & quit monitor
7: end if
8: if navigation response ! = TRUE then
9: node keeps sleep
10: else
11: broadcast navigation message
12: end if
13: end while
Once there is an event occurs, the monitoring message
can be sent by ad-hoc module to find the robot. In order
to save the energy, we set a threshold for the event.
Therefore, the working static nodes are not always at
“working status”.
4. The Mobile Sensor Nodes
The mobile sensor nodes afford the task that detect ad-
versary or dangerous for there is no prior knowledge
about the field or map. Actually, mobile sensor nodes are
the little robots. The working style of mobile nodes
adopts a kind of active learning [9,10].
Since the map of detected field is unknown, the mo-
bile nodes scan the whole detecting area in raster fashion.
A penalized estimate can be constructed. It isc
f

. When
they find these area or fields, they do the estimate of the
area by using average smooth. Then, in this section, the
boundary of dangerous area can be located roughly.
Consequently, the mobile nodes calculate to smooth
these detected regions and get refine localization by av-
erage algorithm. A complexity estimate will be got, just
like the first coarse scan. It isr
f

. Finally, the estimate of
the fields can be generated by coarse scan and refine
localization.
The mean square error can be described as follow, by
using Craig-Bernstein inequality [11]:
1
2''
(1)
12
''
log log
2
dJJd
nn
Eff KKn
n



 
 


 (1)
f

is calculated estimate value that conclude the
coarse estimate c
f

and refine estimate r
f

. K1 and K2
are the constants larger than 0.
Figure 3 shows the process about mobile sensor nodes
detecting and their final localization.
Once the localization of adversary or dangerous re-
gions are estimated, the mobile can communicated each
other, move and stand around the edge of these regions
according average criterion. From then on, the mobile
sensor nodes work as static nodes and send warning mes-
sage. Algorithm 2 has shown the mobile nodes motion
and positioning method.
Algorithm 2 Mobile nodes define dangerous fields
1: while only received the static nodes’ signal do
2: Doing raster scan
3: Collect the sample
4: Launch estimate algorithm
5: if routine = BLOCKED then
6: if other mobile nodes! = stay then
7: stay as a static node
8: sending the warning message
9: else
10: record the sample
11: end if
12: else
13: raster scan
14: end if
15: end while
Thus, the mobile nodes can gather around the edge of
adversary or dangerous. According the adaptive statistic
(a) (b)
Figure 3. The mobile nodes (a) Raster fashion moving; (b)
Mobile nodes location.

S. C. YAO ET AL. 270
algorithm, the mobile nodes can find the adversary fields
as soon as possible. The path length detect can be de-
scribed as [12]:
2
1
(1)
~( )
d
dd
lOn

 (2)
where l is the path length, ~ means the order of polyno-
mial, and d is the dimensions of function. And the time
can be estimated as:
1
(1 )
~( )
dd
tO 
 (3)
After the algorithms of static sensor nodes and mobile
nodes are designed, we can go into the next step. That is
the navigation of robot.
5. Navigation of Robot
In navigation, the robot is Pioneer Ⅱ-AT four-wheel
drive. The robot also has a variety of expansion power
and I/O ports for attachment and close integration of ad-
ditional sensors and other accessories. Expansion includes
an addressable I/O bus for up to 16 devices, two RS-232
serial ports, eight digital I/O ports, five A/D ports, PSU
controllers and more—all accessible through a common
application interface to the robot server software, P2OS
[13]. Therefore, we use the Mica as a communicate de-
vice and serial ports send dictate to the robot.
Most navigation is based on the detect fields map,
which is available in advance. So, the location of robot
can be decided by coordinate of the map [14]. And the
optimal navigation algorithm can be explored. There
have been many algorithms in these fields. But they can
hardly adapt the non-mapping condition. That is to say, if
there is a novel environment or a region that its condition
is difficult to get to know, what these algorithms do is
detect the whole region and acquire the whole informa-
tion of the region. Only after the information of region is
known, can the optimal algorithm can be adopted.
Therefore, the energy may be cost in acquiring informa-
tion. In order to solve the problem, we explore a kind of
non-mapping navigation algorithm to guide the robot to
arrive the event occurring place.
Generally speaking, navigation can be described as 3
questions. That is where I am, where I should go, and
how to arrive [15]. We adopt a kind of dynamic cluster
navigation algorithm. It can decide the location of robot
and use both static nodes and mobile nodes to guide the
robot to arrive the event place.
1) Where I am (Decide the location of robot)
Since what we discuss is no prior field map in this pa-
per; the first thing is the location of robot should be de-
cided. It is the important step.
In order to acquire the location of robot, we adopt a
cluster-navigating algorithm. It depends on the ad-hoc
relay communication. The static nodes work as event
detectors. And several closest static nodes compose as
cluster. The node that finds the event occurs is the leader
of one cluster and broadcast the message. The nodes,
which receive the message, can compose another cluster
and elect a leader, continue broadcasting. So, there is a
cluster heads in every cluster. They can communicate
with the closest nodes, and the nodes are clusters too.
When the event happens, the node, which is closest to
the event, sends a signal both to the next cluster heads
and the nodes in its cluster. At the same time, the hierar-
chy of this cluster is set as 0, the hierarchy of next cluster
heads is 1, and so do as other nodes in the next cluster
until the robot receive the message [16]. Robot receives
the hierarchy of the static nodes. Therefore, the robot can
decide approximately how long the distance between
event and itself not precise. It is enough. And the robot
should distinct the message of event or relay. It can de-
cide itself whether arriving the place or not. At the same
time, robot should also distinct the static or mobile nodes,
so it can decide it is close to danger fields or not.
If the static nodes are too sparse to decide the robot in
a field, it is difficult for robot to make a decision its di-
rection to move. The robot cannot decide where to go.
The system must take measure to make up. Those nodes,
which are not on the key point, should be waken up the
sleeping nodes around them and form a new cluster in
order to inform the robot.
Figure 4 shows the process of detect the robot location.
The detail algorithm of finding robot location is
showed as follow:
Algorithm 3 The Transmission of Event
1: while robot ! = arrived do
2: select working static nodes as cluster
3: if the nodes is too sparse then
Figure 4. The process of finding robot by using cluster com-
munication.

S. C. YAO ET AL.271
4: wake up the sleeping nodes
5: else
6: static nodes work as cluster
7: end if
8: if event = TRUE then
9: hierarchy = 0
10: broadcast message
11: else
12: if received message = TURE then
13: hierarchy < - hierarchy+1
14: broadcast
15: end if
16: end if
17: end while
2) Where should I go
In a large scope, robot can move to the direction
whose hierarchy is always lowing. When robot arrives in
the range of cluster, the static nodes should shut down
the message broadcasting and its hierarchy minus one.
While in the cluster, robot always finds the leader of
this cluster.
3) How to arrive
The robot can follow a routine, which the nodes’ hier-
archy is lowing unless it receives the warning message of
the mobile nodes. When it receives warning message of
the mobile nodes, it have to use the smooth routine in
order to pass the danger field until the warning message
vanish.
Finally, the robot will arrive the place where events
occur when it receive the event’s source message.
The distance from robot to dangerous fields can be
calculated by potential [17]:
1
22
2
11
(, ),0
()
xy R
xy

 

＞ (4)
where R is the Euclidean distance from the edge to the
mobile nods. In this paper, it can be shown as:
2
()(
iieie
2
)
x
xyy   (5)
(xe, ye) is the location of dangerous fields. The energy
of mobile nodes can be defined as follow:
2
1
11
()
22
m
iiiv
i
Vk kv 


 
2
i
(6)
Therefore, we can conclude the algorithm of robot
navigation. There are still some handicaps (they are the
points not area) in the robot routine, but there is sonar in
the robot. The robot can avoid these handicap points. It is
the algorithm 4 shown as follow:
Algorithm 4 Navigation of Robot
1: While event message received = TURE Do
2: if arrived ! = TURE then
3: if mobile warning message ! = TURE then
4: use sonar (or communicate with the static nodes)
follow navigation routine
5: else
6: adopt the smooth routine
7: end if
8: else
9: send arrive message
10: inform the event node turn off requirement
11: end if
12: end while
Since there is no map in advance, what the robot does
is receive the message and do probing motion. Thus, the
time to arrive the event is not shortest and navigation
routine is not the optimal routine. It is the cost of none
prior map available and the lower cost of both static and
mobile nodes. On the other hand, there are no so many
working nodes in this system, the energy cost of nodes
can be saved.
4) How to wake up the sleeping nodes
When the static nodes, which are on the key point, is
too sparse to accomplish the navigation task, it is neces-
sary to wake up the sleeping nodes.
At first, the cluster head, the static nodes on the key
point, acquires the message and sends the message to
next closed cluster head. If the closed cluster head can
receive the message, it will send back a message to the
sending cluster head. That means that the statistic nodes
are not too sparse there. The message can relay to the
next static nodes. If the closed cluster head cannot re-
ceive the message, it cannot send back the message to the
sending cluster head. It means that the statistic nodes are
too sparse there. That is to say, the sleeping nodes in the
cluster have to be waken up in order to relay the mes-
sage.
The construction of the information deliver is shown
in Figure 5.
The static nodes in this system adopt tmote Sky [6]
made by Moteiv Corporation. They are ultra low power
wireless module for use in sensor networks, monitoring
applications, and rapid application prototyping. In an open
Figure 5. The construction of information deliver in a cluster.

S. C. YAO ET AL. 272
field, the communication distance of tmote nodes can
reach 100 meters. But in this system, the communication
distance of tmote nodes is restrained to 10 meters in or-
der to saving energy. In a cluster, there is one tmote node
working as the cluster and the other nodes working as
sleeping nodes. The event occurs, such as the tempera-
ture is over the threshold, the cluster head sends the
message to the next cluster head. If the next cluster head
receives the message, it sends back a message to the
former cluster. Consequently, message can be relayed on.
If the sending cluster head does not receive the backing
message of the next cluster head, the sending cluster
head will send a signal to the sleeping nodes in its cluster,
in order to wake up these sleeping nodes. That is to say,
the low power costs’ nodes (sleeping nodes) will transfer
from the low power costs’ state (sleeping) to working
state (waken up). Once the sleeping nodes are waken up,
they can afford the message delay events.
After the works have been accomplished, we simulate
the sensing and navigation of hybrid sensor network by
using. Figure 6 and Figure 7 is result of simulating ex-
periments.
It can be seen that the energy saving of mobile work is
better than the static network. While, the energy saving
of hybrid network is much better than the mono-static
and mono-mobile network.
The practical experiments are performed with the
setup shown in Figure 8. A sequence process of naviga-
tion is shown from (a) to (d). An emergent event occurs
by a node(the temperature is over the threshold), the ro-
bot (Pioneer Ⅱ-AT) should find the node and arrive the
place in time. Part (a) shows the initial experiments
situation. There are 100 tmote nodes (static nodes) and 8
small robots with the tmote nodes on them(mobile nodes)
in a 25m2 (5m×5m) filed. The lawn simulates the dan-
gerous fields, and the scope can be detected by moisture
Figure 6. The result of simulating and the trajectory of robot.
Figure 7. The energy conservation of different network.
(a)(b)
(c) (d)
Figure 8. The practical experiments of the navigation.
sensor on tmote nodes. Therefore, the range of dangerous
fields can be defined by them. Consequently, part (b)
shows the dangerous fields are defined. Part (c) shows
the process of navigation, and part (d) shows the robot
arrive the emergency place (red circle) finally.
Despite the difficulties with the dangerous fields’
definition, the navigation is successfully performed many
times. Although arriving the final location of the robot is
not so quickly (about 3 minutes), the robot avoids the
danger fields and being guided by this hybrid sensor
network system indeed. We plan to conduct further re-
search to get higher speed of this navigation system.
6. Conclusions
This implementation demonstrated robot guidance by a
hybrid sensor network. The energy saving in sensor net-
work is an important problem. We take it as the primary

S. C. YAO ET AL. 273
problem in this system. The skeleton of adversary or
dangerous fields can be outlined by using the mobile
sensor nodes and the robot navigation comes into reali-
ties on the ad-hoc method. In spite of the less optimal
routine, it is an effective scheme when there is no map of
the whole detecting fields in advance.
However, the precision of the navigation system
greatly depends on the robot hardware. Navigation can
be problematic if environmental factors are not so ideal,
such as electrical cables, higher power electrical equip-
ment and steel structures exist. Although it is possible to
compensate for these effects in a known environment, a
search situation may not permit elaborate offline calibra-
tion. In addition, the response speed of whole system
should be raised higher. We are going to do further re-
search on them.
7. Acknowledgment
We would like to give special thanks to Dr. Jindong Tan,
the professor of Michigan Technological University in
U.S.A. for his providing of experimental equipment and
environment. The same thanks go to Qi An, who helped
me with studies of the project and sensor network,
TinyOS.
8. References
[1] D. Moore, J. Leonard, D. Rus and S. Teller, “Robust
Distributed Networklocalization with Noisy Range Meas-
urements,” Proceedings of the 2nd International Confer-
ence on Embedded networked sensor systems, Baltimore,
2004, pp. 50-61.
[2] R. Szewczyk, A. Mainwaring, J. Polastre and D. Culler,
“An Analysis of a Large Scale Habitat Monitoring Ap-
plication,” Proceedings of the 2nd International Confer-
ence on Embedded networked sensor systems, Baltimore,
2004, pp. 214-226.
[3] J. Polastre, J. Hill and D. Culler, “Versatile Low Power
Media Access for Wireless Sensor Networks,” Proceed-
ings of the 2nd International Conference on Embedded
networked sensor systems, Baltimore, 2004, pp. 95-107.
[4] M. Batalin, M. Rahimi, Y. Yu, D. Liu, A. Kansal, G. Suk-
hatme, W. Kaiser, M. Hansen, G. Pottie, M. Srivastava and
D. Estrin, “Call and Response: Experiments in Sampling
the Environment,” Proceedings of ACM Sensor Systems,
Los Angeles, 2004.
[5] M. Rahimi, R. Pon, W. J. Kaiser, G. S. Sukhatme, D. Estrin
and M. Srivastava, “Adaptive Sampling for Environmental
Robotics,” Proceedings of IEEE International Conference
on Robotics and Automation, Los Angeles, 2004.
[6] http://www.moteiv.com
[7] http://www.crossbow.com
[8] R. Willett, A. Martin and R. Nowak, “Backcasting:
Adaptive Sampling for Sensor Networks,” Proceedings of
IEEE International Conference on Information process-
ing in sensor networks, New York, 2004, pp. 124-133.
[9] R. Castro, R. Willett and R. Nowak, “Faster Rates in Re-
gression via Active Learning,” Proceedings of Neural In-
formation Processing Systems, 2005. http://homepages.cae.
wisc.edu/ rcastro/ ECE-05-3.pdf
[10] M. Horstein, “Sequential Decoding Using Noiseless Feed-
back,” IEEE Transactions Information Theory, Vol. 9,
No. 3, 1963, pp. 136-143.
[11] J. Borenstein and H. R. Everett, “Navigating Mobile Ro-
bots: Sensors and Techniques,” John Wiley & Sons, 1992.
[12] M. Burnashev and K. S. Zigangirov, “An Interval Esti-
mation Problem for Controlled Observations,” Problems
in Information Transmission, Vol. 10, 1974, pp. 223-231.
[13] O. Khatib, “Real-Time Obstacle Avoidance for Manipu-
lators and Mobile Robots,” The International Journal of
Robotics Research, Vol. 5, No. 1, 1986, pp. 90-98.
[14] J. Cortes, S. Martinez, T. Karatas and F. Bullo, “Cover-
age Control for Mobile Sensing Networks,” IEEE Trans-
actions on Robotics and Automation, New Orleans, April
2004, pp. 243-255.
[15] A. Korostelev, “On Minimax Rates of Convergence in
Image Models under Sequential Design,” Statistics and
Probability Letters, Vol. 43, 1999, pp. 369-375.
[16] G. Golubev and B. Levit, “Sequential Recovery of Ana-
lytic Periodic Edged in the Binary Image Models,” Mathe-
matical Methods of Statistics, Vol. 12, 2003, pp. 95-115.
[17] R. Castro, R. Willett and R. Nowak, “Faster Rates in Re-
gression via Active Earning,” Proceedings of Neural In-
formation Processing Systems, Vancouver, 2005. http://
homepages.cae.wisc.edu/rcastro/ECE-05-3. pdf