Proceedings 2005 IEEE International Symposium on

Computational Intelligence in Robotics and Automation

June 27-30, 2005, Espoo, Finland

Path Planning Protocol for Collaborative Multi-Robot Systems

Jenhui Chen and Li-Ren Li
Department of Computer Science and Information Engineering
Chang Gung University, Kweishan, Taoyuan, Taiwan 333, R. O. C.
[email protected]

Abstract- Wireless ad hoc sensor networks are being devel- these proposed model need the exact model of the environment
oped to collect data across the area of deployment. These tech- and is difficult to handle correctly dynamic modifications of
nologies enable multiple robots to form a temporary multi-robot the the environment due to the addition of objects and presence
team and cooperate with each other to launch a complex mission.
A path planning algorithm and well-organized communication of obstacles with mobility.
protocol are needed when the multi-robot systems have to search On the contrary, some researches about the unknown en-
for or reach a designate target. It is more complex in designing vironment and single robot scenario [6], [11], [24], UEPP
a collaborative path planning algorithm and communication incorporates and reflects the sensing information to a robot's
protocols for multi-robot systems since it has to consider avoiding planning process as apposed to classical motion planning.
intra-team collisions, energy efficiency, information sharing and
cooperation problems, etc. Moreover, unlike single robot path Therefore, an absolute localization is not required between the
planning problem, a multi-robot system is usually constructed robots and the environment. In these circumstances, the robots
by several simple, cheap, function-restricted, and energy-limited has to acquire through its sensory inputs a set of stimulus-
robots to plan a path toward the target by cooperative fashion. response mechanisms instead of a structural modeling of the
This is the main advantage of the puny multi-robot system. environment. In this scheme, the robot is generally expected
Therefore, in this paper, we propose a simple but efficient collab-
orative path planning algorithm (CPPA) and a communication to carry out simple tasks. Nevertheless, these methods are
protocol for the sensor multi-robot systems where the energy only operated in a single robot system. The reason of using
consumption is reduced as well as the duration of reaching multiple robots is to prevents the robots break down in their
the goal is shortened. Moreover, considering the survivability mission-executing for increasing the survivability. There is
of the mission, the proposed algorithm can enable the sensor also a drawback about energy that every robot needs to do
robots to complete the mission even if some robots are failed by
accidents. Experiment results show that the energy consumption the same calculations and decision-making in the single robot
and computation of proposed algorithm is lower than the multiple system. Although many methods of path planning in unknown
independent robots or other methods. environment have been proposed for multi-robot system [1],
[3], [5], [25]. They seldom consider the power saving as well
I. INTRODUCTION as the complexity of the method in this system. Therefore,
There are many popular researches about sensor networks we propose a power-efficient path planning protocol named
in recent years [10], [14], [22]. The character of sensors that it collaborative path planning algorithm (CPPA) for multi-robot
can detect the information of the specified regions where we to operate easily in unknown environment. The decision of
are interested like humidity, brightness, degree of shake, and path planning in CPPA is decided by a coordinator of a
so forth [2], [7]. Besides, with the advances in wireless ad hoc cluster with the environmental information sensed by other
communication and robotics technology, it can be possible to member robots. With the cooperation of the coordinator and
organize teams (swarms) of the autonomous multiple sensor members, the cluster can reduce the communication to reach
robots to finish complex missions that comprise several teams their destination.
[9]. These sensor robots generally cooperate with each other The remainder of paper is organized as follows. Section II
to complete an assigned mission and use collaborative strategy describes the related technologies and architecture of multi-
for power saving, increasing the survivability, and improving robot system. In Section III, we propose a concept of the
the successful probability of the mission, etc. protocol standard with minimap to help the decision-making of
The path planning problems had been discussed for several the coordinator. Then, in Section IV, we present our proposed
decades, because that this problem is an essential research CPPA for implementation of this protocol. We perform a
issue in robotics since robots need to plan a path to reach series simulation models to evaluate the proposed algorithm in
a goal. General speaking, the path planning problems are Section V. Finally, we give some conclusions in Section VI.
classified as two categories: known-environment path planing
(KEPP) and unknown-environment path planning (UEPP). The II. COLLABORATIVE MULTI-ROBOT SYSTEM
KEPP methods assume a complete knowledge of both the Fig. 1 shows a sketch of multiple sensor robots scenario
robots and the environment [8], [19], [23]. Its main advantage searching for a target in an unknown environment with many
is to prove the existence of a solution that permits the robot obstacles and needing a collision-free path planning algorithm
to reach its destination and to generate collision-free map- to go to the target. These mobile robots equips several sensors
making. However, they have some well-known drawbacks that and wireless communication equipments. There is no global

0-7803-9355-4 / 05 / $20.00 ©2005 IEEE. 721

 Unknowti
Fnv ironrnemtS

v
I
Robo)t
Robot
..
'/

1,J
*<1
t
1
i

j, z-iJ RtbotRO

( )oordinator

Fig 1. An illustration of the path planning ill unknown ensironment with

multiple sensoi robots
Fig. 2- The example of S x 8 fixed stiuctural mninaimap with obsticles f(t a
robot in unknown env ironment.

positioning systeni (GPS) since all sensor robots are in indoor columns e.-ngith
environment. and needs indoor position estimation methods
Sourcc_ID Numr
to indicate the position. Most existing localization algorithms
make use of trilateration or multilateration based on range Dcstination ID Num 4 bits
measurements obtained from received signal strength indicator X-Coordinate sa0 bits
(RSSI) [4] time of arrival (ToA) [20]. time difference of
arrival (TDoA) [ 18], and angle of arrival (AoA) [17j]. The RSSI Y-Coordinatc I 0) bit
technique is employed to measure the power of the signal at Enviommental Data vai- bits
the receiver and, because of its simplicity, it is considered in
our work. FigJ 3. The packet formnat of the minimap sending trom the muembers to
These sensor robots are organized as a team and one of them their coordinator.
plays a coordinator. The others serve as members and they
communicate with each other by wireless devices. The coordi-
nator is like a guider to lead their members to go to the target. bound h,X and the low bound 1). Each robot, after receiving
The communication protocol is IEEE 802.15 4 [12], which is the beacon, will judge the -il, whether smaller than the hl, If
a low cost and low energy consumption equipment for short 'hl < h1, the robot R- will turn it direction and goes toward
range (-- 10 meters in indoor environment) communications. the coordinator until the 'j is larger than or equal to the 1-,
In the aspect of the direction measurement of the robots, each
sensor robot equips a electronic compass, which can support IIIt MINIMAP INTEGRAIION
a global direction of the robot. to indicate its current direction At the Section II, we have known that tihe coordinator
[2 1]. With the electronic compass, the sensor robots can know collect the information of the environment fromt its members
which direction they lace. The direction of the robot d is easily for maintaining the global map information to help itself
represented as an angular magnitude by electronic compass. do decision-making of path planning. Therefore, we should
Besides, the inter-distance of robots in a cluster should be know the packet format of the data communication between
maintained to avoid collisions or wandering away. We consider the coordinator and the members. So in this section, we
a cluster consisting of n. sensor robots Rf. R2*. Rt and Dij try to make the sensing data of environmnent a fixed packet
represents the distance between two sensor robots fi and RJ. format. There are two packet format of communication for
The robot collision avoidance problem can be considered same the coordinator and the members. One packet format is for the
as the obstacle avoidance by using several techniques such member to send information of environment to its coordinator.
as infrared [15] and ultrasound [16], etc. These techniques and the other is for the coordinator to request the members to
are designed to overcome the collision problems but not gather their information of the environment. With this format
wandering away problem. We use the RSSI value ij to estimate the robots can more easily do communication with others in
the distance between two robots [13]. The RSSI value can be a common protocol.
obtained by received communication packet from other robots. First, we define a fixed 8 x 8 minimap model to represent
If the measured signal strength i1, from the received packet the environmental data of a robot in its coverage. In Fig. 2
exceeds that of the robot Rf by a threshold ht. we say that the a robot fills the grids with the color (environmental data
robot Rf is in the lower boundary range. Otherwise, the Rf =1) if it sense the obstacle in it. The robot can adjust
is wandering away from the coordinator and should be pulled the size of na x on minimap according to the number of
back. To maintain the distances between the coordinator and sensors the robot have. Therefore, we make the itiforrmation of
its members, the coordinator periodically broadcasts a beacon minimap into a fixed packet with 64 (8 x 8) bit streams. For
B with its coordinate to its members. To avoid the ping-pong example, the 64 bits environmental data of the robot in Fig. 2
effect of using the threshold, we set two thresholds the upper are "00000000, 01l11100, 00000000, 00000000, 00000000.

722
 The Coordinator The
nironm, t
The.
member 3
The members gather the
information of environment and
send a minimap to the coordinator The
member 4
- Sending the packet format in Fig.5
- _ Sending the packet format in Fig.3

Fig. 6. The sketch of communication model between the coordinator and its
members.

Fig. 4. The members sense the information of environment integrated by number of member needs to gather the information of the
the coordinator and decide all the free-space ways.
environment. If the message of coordinator needs to broadcast
columns length
to all the members in its cluster, the column of the Destina-
tionlIDNum is contain "11 11" (reservation for broadcasting).
::Soure_ID_ Num 4 bits There are three commands in the Command-Type column of
the packet that we can categorize with its command actions.
-DestinationD Numn 4 bits
The Command-Type = 00 represents the broadcast message
Command Type: 2 bits (or beacon) with unchanged direction of its cluster, and the
-::t -:-::X-Coordinatei: :: :i;:0-0: :: 10 bits Command-Type = 01 is also the broadcast message but with
changed direction of its cluster. The last, the Command-Type
Y-CoordinateV 10 bits = 11 is used by the coordinator to gather the information of
Compasis Direction: 9 bits environment from the specified id robots in the column of the
DestinationID-Num. And the column of X-Coordinate and Y-
Fig. 5. The packet format of the request sending from the coordinator to its Coordinate are the coordinator's coordinate. The last column is
members. used for containing the degree of the coordinator's angle by its
electronic compass. Using these two fixed packet format, the
coordinator and members can easily request for and transmit
01000000, 01000000, 00000000"' respectively. So we can with the environmental information to each other.
make the sensing data a packet format in Fig. 3. This packet
format is used for the members to sending the information IV. INTRA-TEAM COMMUNICATION PROTOCOL
of environment to the coordinator. In this packet format, the By the integration of the minimaps, we can clearly construct
column of SourcelID-Num and DestinationID-Num are filled the communication model in Fig. 6 between the coordinator
with the id number of the member itself and its coordinator. and the members. Therefore, in this section, we describe
And the column of X-Coordinate and Y-Coordinate contain its the proposed collaborative path planning algorithm (CPPA)
coordinate. The environmental data are extended into the last with intra-team communication protocol in detail. Taking
column of the packet with adjustable fixed bits. Fig. 4 for example, there are nine sensor robots (n = 9)
After defining the minimap model, the robots can easily gathered as a cluster to cooperate with each other to go to
send and the coordinator can correctly recognize its sensing a given target. First, R, receives the (xt,yt) of the target
information by the standardized packet format. The Fig. 4 and calculates the angular magnitude to the target and to
is the sketch of the environmental shape integrated by the turn its direction to the target. The R, then broadcasts its
coordinator in path planning. The coordinator receives all members a request packet including the target's coordinate
standardized minimaps of its members where it is interested (Xt,Yt), its coordinate (x,,yc), and the facing direction to
to build the partial view of the environment to check all the request its members to collect related map information. The
space-free ways for decision-making. Therefore, the packet member Ri, i e {I,... , n - 1}, after receiving the request
format for the coordinator is in Fig. 5. In this packet for- packet, will collect the related environment information and
mat, the SourceiD-Num is filled with the id number of the reply its minimap including the shapes of obstacles if any and
coordinator, and DestinationIDNum is filled with the id its coordinate (xi,yi) back to R,.

723
THE COORDINATOR ALGORITHM THE MEMBER ALGORITHM
BEGIN BEGIN
turnleft_counter +- 0 far-away +- 0
turm-right_counter +- 0 while not reaching the Target do
set Command-Type i- 11 receive the broadcast information ((x,, yc), v)
v = f ((xc , yc), (xt, yt)) if far-away flag is set and ? is not bigger than hu then
broadcast ((xc, xc), v) V = f ((Xm, Pm), (Xc, yc))
while not reaching the Target or receive the Mm do else
if detecting an obstacle from sensor robots (x0, yo) then if r is less than h,
if OBSTACLEDECISION(x0, yo) then set the far-away flag
if turnmrightLcounter flag is set then v = f((xm, ym), (Xc, Yc))
v' = counterclockwise-choose-free-space (xc, yc, v) else
else V =vc
if tumleft-counter flag is set then clear the far-away flag
v' = clockwise.chooseifree-space (xc, yc, v) constructing the minimap
else send the minimap back to its coordinator
v' = choose_free.space (xc, yc) if OBSTACLEJDECISION(x0, y,) then
if v' is no more than v and tumleft-counter is not set then v, = f((xm, ym), (xc, Yc))
set the turnmeft.counter flag if v' is no more than v then
else HI v' is more than v and tumrnightLcounter is not set v= clockwise-choose-free-space (xm, ym, v)
set the turn_right_counter flag else HI v' is more than v
set Command-Type +- 01 v = counterclockwise-choose-free-space (xm, ym, v)
else endwhile
clear the turnileftLcounter and turn.right_counter flags END
set Command-Type +- 00
stack ((Xc, yc), v) - ((Xc, yc)', v') Fig. 8. The algorithm of the member process.
broadcast ((Xc, yc), v)
endwhile
END maintain its minimap and avoid the obstacles they meet. The
Fig. 7. The algorithm of the coordinator process. communication overhead is also reduced since members only
communicate with the coordinator when they meet the obstacle
in its straight direction.
The coordinator will wait for a while or its countdown
expired to collect all the minimaps from its members to build V. EXPERIMENT RESULTS
up a temporary view of the environment for making decision. To illustrate the efficiency of the proposed CPPA on the
According to the (xi,yi) and the fixed structural model of path planning of multi-robot systems, we perform a series
minimap, the coordinator can determine which of locality of simulation scenarios to evaluate the performance of the
Ri is. The coordinator combines different minimaps from its CPPA. We perform four different scenarios, shown in Fig. 9, to
members, it can build a temporary map as shown in Fig. 4. evaluate the CPPA. These simulation scenarios have individual
The RC then, according to the v and several possible space- characters, such as cross-shaped obstacles, double frame-
free ways, chooses the nearest (with high priority) space-free shaped obstacles, and other easily trap-making mazes. Initially,
way to the target to make a smarter decision. ten mobile sensor robots are randomly distributed in the left-
The direction will be changed while the team meets an down corner in the simulation area. One of them serves as the
obstacle. Since the CPPA is based on cooperative fashion, the coordinator and the others serve as cluster members. The target
members will send the minimaps to the coordinator if they is assumed in the right-up corner of the area. The simulation
meet obstacles. Thus, when Ri runs into a obstacle in a straight parameters are shown in Table I.
direction, it will announce its minimap to RC for the decision
making. When RC receives the address of a obstacle, it will TABLE I
update its map information and determine whether they have SYSTEM PARAMETERS IN SIMULATIONS
to change a new direction or not. If RC decides to change
Simulation Parameter Normal Value
a new direction, it will broadcast this new information to Radio data rate 250 kb/s
its members with the new direction (set Command-Type = Radio transmission range 10 m
01) and its current coordinate (xc,yc). Thus, members will Radio transmission power 1400 mW
move toward the new direction announced by RC if they have Radio receiving power 1000 mW
Radio idling power 830 mW
exceeded the (xc, Ye). If RC does not change a new direction Radio sleeping power 130 mW
(the column of Command-Type is still 00), Ri will avoid the CPU clock rate 400 MHz
obstacles and get closer to RC when its new direction has no Computation power 500 mW
Ultrasound sensing range Im
obstacles. The algorithms of the coordinator and members are Simulation map length 1000 m
shown in Fig. 7 and Fig. 8, respectively. Simulation map width 1000 m
These processes are proceeded interactively until they reach
the goal.To alleviate the energy consumption, the direction to For the simplicity of representation, we use a black point to
the target is computed by coordinator only and the members represent the cluster. The movement of the cluster is presented
follow the direction to go to the goal. They only needs to by using series points. Fig. 10 shows the experiment results

724
 T

T booo.
- 0 0

rrl
--Single robot
-A
-----CPPA
5500

o0 5000

So0 R
~P ~4500
scenario 1 (sl) scenario 2 (s2)
0.
~ 4000

3500

WM 1. I 5 6
-
7 ~8 9q
Number of robots~

scenario 3 (s3) scenario 4 (s4) Fig. 11. The comparison of the multi-robot by CPPA and single robot in
the designed sl scenario.
Fig. 9. Four scenarios of the simulation.
S000OLI II

-- -- Singl robot

5500I

5000

4500

scenario (sl) scenario 2 (s2)

4.000 )l
.3500 f

a30oo(' .I
6
1.
7 8
~~~ 9: 10T
Number of robots

Fig. 12. The comparison of the multi-robot by CPPA and single robot in
scenario 3 (s3) scenario 4 (s4) the designed s2 scenario.

Fig. 10. The moving path of a cluster of multi-robot by using CPPA.

and movement. We try to considerate all the possible energy

consumptions in real environment to make the simulation more

by our simulation in the four different scenarios. We can see

realistic. Fig. 1, Fig. 12, Fig. 13, and Fig. 14 show the average
that the proposed CPPA leads the robots toward the target
energy consumption of each robot by the CPPA and single
smoothly. Moreover, in scenario 3 and 4, we can see that, base robot in the four scenarios, respectively. We can see that the
on the CPPA, initially the coordinator plans a wrong path to the
power consumption of each robot by using the CPPA is more

target by decision making. However, the CPPA will collect the lower than the single robot system when the number of the
environment informnation to adapt their moving path to reach cluster is increasing. The power consumption is getting lower
the target correctly. when the number of robots in a robot team increases. This is
To compare the energy efficiency of the CPPA with other because that CPPA adopts cooperative strategy to plan a path
single robot path planning algorithms. At last simulation, we
to the given target and thus reduces the power consumption
try to make a comparison of average energy consumption by one single robot.
with a single robot. The energy consumption in the CPPA
VI. CONCLUSIONS
includes decision-making computation (only in coordinator),
wireless communication (transmission, receiving, idling, and In this paper, we propose a power-efficient path plan-
sleeping), and movement, but the energy consumption in gen- ning protocol named collaborative path planning algorithm
eral single robot only includes decision-making computation (CPPA) for a multi-robot system without global positioning

725
 aoOOO
[21 P. Bergamo et al., 'Collaborative Sensor Networking TOwards Real Tinie
L-'---
-A-I Single
0PPA
robot
Acoustical Beamforming in Free-Space. and Limite Reverberanceg
8500 -
IEEE Twscins. Mobile Computing. vol 3. no. 3 pp. 211 -224. July 2004
[3] S. Boonphoapichart S. Komada, T. Hori. and W.A. Gruvexi "Robot
3 8000 Motion Decision-macking System in Unknown Environments'" in Proc%
_5 IEEE ICRA 2003. vol 3, pp. 4197-4202, Taipei, Taiwan, Sept, 2003.
--£ 4t [4] N Bulusu. ..Heidemann, and D. Estn, "GPS-less Lowcost OutdoJor
z
-9 A-L
C- 7500 h-
Localization for Very Small Dex ices." [FF
IEEE onaol OHf1i11_ VsoL
no. 5. pp. 28-34, Oct. 2000.
7000
`j] Z. Cao, L. Xie, B. Zhang. S. Wang, and M. T-an, FrFi- oiatkorn Constrained
5 Ir Multi-robot System in Unknown Environments." .n P.roC. IEEE ICRA=;
6500 r 2003 vol. 1, pp. 735-740. Taipei, Taiwan, Sept. 2_003.
[60 W. Chen, C. Fan. and Y. Xi, 'On-line Safe Path Planning in Unknown
60004 Environments," in Proc. IEEE ICRA 2009, vol 3, pp. 41 9 -1 1496 Taipei.
Taiwan, Sept. 2003.
[7' W.-P. Chen, J.C. Hou, and Lui Sha. 'Dynaimic Clusterinklotfr Acoustic
Target Tracking in Wireless Sensor Networks,' lEE TIras. Mobile
5000 4_--
Comiputing, vol. 3. no. 3. pp. 258-271, July 2004,
6 7 8 9 10 [81 J.L. Diaz de Leon S. and J.H. Sossa A., "Automatic Path Planning t a
Nulmber of roboi Mobile Robot among Obstacles of Arbitrary Shape" IEEE Twanos Sosr,
Mati. Cvbern. B. sol. 28. no. 3, pp. 467-472 Joine 1998).
Fig 13. The comparison of the multi-robot b' (CPPA and slngle robot in
[91 M. Gerla and Y. Yi. "Team Commiiunications among ALtonormous Senisor
Swarms."' ACM SIGMOD Record. vol. 33. no. 1. pp. 29-25. Mar 2004
the designed s3 scenaro(. 101 P. Havinga, J.C. Hou. and F. Zha, "Wireless Sensor Netwxtiks" IEEE
lVireless Commiun.. ol. II. no. 6. pp. 4-5, Dec. 2Q004.
13000 1--' 1
7
1 11 A.hl. Hussein and A. Elnagar. 'A Low Cost Path Planer it Unknowx
- Environnients for Autonomous Mobile Robots." in Pnos IEEEt-t Sywlp
-- Single roboll CIRA-2003. vol. 2, pp. 724-728, Kobe, Japan. July 2003.
o CP-PA
r-

[121 IEEE 802.15.4 Working Group, "Part 154 Wireless Medium \Iccss
"12000 Control (MAC) and Physical Layer (PHY1 Specifications for Low
Rate Wireless Personal Area Networks (LR-WPANsI ANSI/IEEE Std
802.15.4. Feb. 2003.
I -
A0S --A 4- [131 K.-I. Itoh, S. Watanabe. J.-S. Shih. and T. Sato, 'PeIrlfrance of Handotf
Algorithm Based on Distance and RSSI Measurements." IEEE Iraosn
Veh. Te/i.. sol. 51, no. 6, pp. 1460-1468 Nov. 2002.
1000 [tl141 P. Julian. A.G. Andreou. L. Riddle. S. Shamma, D.H. Goldberg, and
10000 -: G. Cauwenberghs. "A Comparative Study of Sound Localization Algo-
rithms for Energy Aware Sensor Network Nodes." IEEE Trans(Circuits
SvSt., vol. 51, no. 4. pp. 640-648. April 2004.
9000 [
I15] V Jungnickel, A. Forck, T. Haustein. U Krugei. V. Pohlc and C. von
Helmolt. "Electronic Tracking for Wireless Infrared Comrmunications."
JEE Trans. Wireless Commun., vol 2. no. 5. pp, 989--999, Sept. 2003.
J [16] P.J. McKerrow and S. min Zhu, "Modelling Multiple Reflection Paths in
8000 4- 1 _ w
_-4P Ultrasonic Sensing," in Proc. IEEE/RSJ IROS-1996, vxi, '. pp. 284-29 "1
4 5 6 7 8 9 10 Osaka, Japan, Nov. 1996
Number of robotsS [17]I7 . Niculescu and B. Nath, "Ad Hoc Positioning System (APSI using
AoA." in Proc. IEEE INFOCOM 2003, vol. 22, no. 1. pp. 1734-1743
Fig. 14. The comparison of the multi-robot 'ov CPP5.A a 'male robolt in [18] N.B. Francisco,
San CA, April 2003.
the desitrned s4 scenario, Priyantha, A, Chakraborty. and H. Padmanabhan. Iihe Cricket
Location Support System." in Proc, A CM MbiCom .()fd. rpp. 32--t3
Boston. MA. Aug. 2000.
[191 E. Sack, "Path Planning for Planar Articulated Robots using Configura-
system (GPS). The energy consumptioi :nis reduced by using tion Spaces and Compliant Motion." IFEE Tanso. Robo6. Automat., sol.
19. nio. 3. pp. 381 390. June 2003.
a structural model of protocol and a c
multiple mobile robots. Comparing to thie signal robot system,
*ooperative
fashion of [20] A. Savvides. C.-C. Han. and M. Srivastava. D'Dynamic Fine-Gramlled
Localization in Ad-Hoc Networks of Sensors," in Proc, ACM MobiCtnn
multi-robot system can enhance not on ythe performanc.e of 2001. pp. 166-179, Rome, Italy, July 2001.
execution but the survivability. Experin nent results also show P[21 bileSuksakulchai,
S. S. Thongchai. D.M. Wilkes, and K. Kawamiura. 'Mo-
Robot Localization using an Electronic Compass for Corridor
that the proposed CPPA gets lower eneIrgy consumption than Environment," in Proec. IEEE Int. Conj. Systemn. Man and Cybernoetis
those robots without teamwork, pp. 3354-3359, Nashville, TN. Oct. 2000.
[22 NI. Sveda and R. Vrba, "Integrated Smart Sensor Networking Framework
for Sensor-based Appliances.' IEEE Setnsors J. vl3o no- 5. pp, 579
ACKNOWLEDGEME: N1 586, Oct. 2003.
This work was supported in part by [23] L.P. Tomas and A.VW. Michael, "An Algorithm for Planning Collision.
free Paths Among Polyhedral Obstacles" Commmn. ACM vol 2. no
Council, Taiwan, R.O.C.. under Contrac,t NSC93-22 13-E- 182- 10. Oct. 1979.
02,2 [24] B. Tosar. S.M. LaValle, and R. Murrieta. '.Optimal Navigation and
Object Finding without Geometric Maps or Localization'" in Prot'. IEEE
REFERENCES ICRA-2003, vol. 1, pp. 464-470, Taipei, Taiwan. Sept. 2003.
[25] A. Zhu and S.X. Yang, "Path Planning of Multi-robot Systems with
[1 ] R. Alami, S. Fleury, M. Herrb, F. Ingran d, F. Robert. "Multi Robot Cooperation." in Proc IEEE Int, Symp. CIRA-200-3 vol. 2, pp. 1028
Cooperation in the MARTHA Project.' IEiFEg RobotAt. 4uomatX Mag. Q1033, Kobe. Japan. July 2003.
vol. 5, no. 1 pp. 36-47, Mar. 1998.

726