Self-Optimizing Algorithms for Mobile Ad Hoc Networks

based on Multiple Mobile Agents

Yasushi Kambayashi

, Tatsuya Shinohara

and Munehiro Takimoto

Department of Computer and Information Engineering, Nippon Institute of Technology, 4-1 Gakuendai, Miyashiro-machi,

Minamisaitama-gun, 345-8501 Japan

Department of Information Sciences, Tokyo University of Science 2641 Yamazaki, Noda 278-8510 Japan

Keywords: Mobile Ad Hoc Network, Mobile Software Agent, Multi-agent System, Multi-robot System, Self-organizing

System, Swarm Intelligence, Optimization.

Abstract: This paper presents algorithms that form optimal connecting configurations for Mobile Ad Hoc Networks

(MANETs). MANET is a computer network that is dynamically formed by autonomous mobile nodes.

Today, the communication network is one of the most important infrastructures. When it is lost by either

natural or accidental disaster, the recovery of the communication network should be one of the first

priorities. We are proposing a way of constructing an extemporized communication network on the spot by

a herd of mobile robots that communicate by wireless link. The networks we are considering are formed by

multiple relay robots; therefore the algorithms are naturally distributed ones and executed by the herd of

relay robots. The relay robots move cooperatively but without any central control. In order to collect and to

distribute enough information to coordinate the behaviours of participating relay robots, we employ mobile

software agents that we have developed and succeeded in using many applications. There are a number of

multi-robot systems that take advantage of MANET, and look for efficient use of relay robot while

maintaining connectivity. Our study contributes this line of investigation. The numerical experiments show

that our algorithms provide optimal configurations in certain cases.

1 INTRODUCTION

In the modern society, the communication network

is one of the most important infrastructures. When it

is lost by either natural or accidental disaster, the

recovery of the communication network should be

one of the first priorities. Under such an assumption

we have conducted a project that constructs an

extemporized communication network on the spot

by a herd of mobile robots that communicate by

wireless link. They are expected to form a Mobile

Ad Hoc Network.

Mobile Ad Hoc Network (MANET) is a

computer network that is dynamically formed by

autonomous mobile nodes. Such mobile nodes are

connected through wireless links without relying on

any central controller or established infrastructure.

The participating mobile nodes can freely and

dynamically self-organize into arbitrary and

temporary network topologies.

The application we have in our mind is

constructing a temporary communication network in

a contaminated area polluted by radioactive

substances or dangerous gas because of natural or

accidental disaster that prevent human activities.

Under such conditions, constructing MANET by

using a multi-robot system should be a natural

choice. It may be desirable for us to connect

arbitrary two points. For example, we may want to

connect the control centre of a nuclear power station

and a reactor with problems by using scattered

mobile robots with minimum costs so that robots can

work as long as possible without human intervention.

A multi-robot system consists of a large number

of homogeneous robots that have limited capacity,

but when combined into a group, they can generate

more complex behaviours (Parker, 2008). In multi-

robot systems, robots communicate with each other

to achieve cooperative behaviours. There are three

major advantages of multi-robot systems over single

robot systems (Stone and Veloso, 2000) (Yasuda

and Ohkura, 2005). The first is parallelism; a task

can be achieved by autonomous and asynchronous

robots in a system. The second is robustness; this is

156

Kambayashi Y., Shinohara T. and Takimoto M..

Self-Optimizing Algorithms for Mobile Ad Hoc Networks based on Multiple Mobile Agents.

DOI: 10.5220/0004818901560163

In Proceedings of the 6th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2014), pages 156-163

ISBN: 978-989-758-016-1

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

realized through redundancy. The system can have

more robots than required for a certain task. The

third is scalability; a robot can be added to or

removed from the system easily. We have taken

advantage of these properties.

We have implemented several multi-robot

systems such as cooperatively assemble themselves

at energy-wise optimal locations (Kambayashi et al.,

2012), and serialize themselves (Shintani et al.,

2011). For all the multi-robot systems, we have

designed and implemented multi-agent systems that

control the robot systems. A control system based

on multiple software agents can control robots

efficiently. Multi-agent systems introduced

modularity, reconfigurability and extensibility to

control systems which had been traditionally

monolithic. It has made easier the development of

control systems on distributed environments such as

multi-robot systems.

On the other hand, excessive interactions among

agents in the multi-agent system may cause

problems in multi-robot environments. In order to

mitigate the problems of excessive communication,

we have developed mobile agent methodologies for

distributed environments (Kambayashi and

Takimoto, 2005). In a mobile agent system, each

agent can actively migrate from one site to another

site. Since a mobile agent can bring the necessary

functionalities with it and perform its tasks

autonomously, it can reduce the necessity for

interaction with other sites. In the minimal case, a

mobile agent requires that the connection is

established only when it performs migration (Binder

et al., 2001).

We have achieved energy saving multi-robot

systems through multiple mobile software agents

that migrate in a herd of mobile robots to collect

information about them, as well as drive the

minimum number of them based on the collected

information. Moving software agents instead of

physical robots greatly save energy consumption.

In this paper, we propose a multi-robot system

that employs MANET through which software

agents migrate. By using the software agents, the

relay robots in MANET can cooperatively

coordinate themselves into optimal locations to

make the shortest communication route with a

minimum number of relay robots. We propose two

algorithms to form a optimal route via relay robots,

and discuss the pros and cons of the two.

The structure of the balance of this paper is as

follows. In the second section, we describe the

background of our research. In the third section, we

present the two algorithms to form optimal

configurations. In the fourth section, we present the

numerical experiments through simulations to

demonstrate the effectiveness of our algorithms and

discuss our observations. In the fifth section, we

conclude our discussions and suggest future work.

2 BACKGROUNDS

There are a number of multi-robot systems that take

advantage of MANET. Heo and Varshney

considered the sensor coverage problem for the

deployment of wireless sensor networks (Heo and

Varshney, 2003). They have proposed a distributed

algorithm for the deployment of mobile nodes, not

necessary autonomous robots, to cover a certain

region by limited number of nodes and limited

communication range. They focus on the sensor

coverage problem and did not discuss the multi-hop

relay problem of ours.

Voyles et al. introduced a new multi-hop

protocol (Voyles et al., 2009). As we have done,

they used Bluetooth ad hoc wireless communication

for use in sparse, highly volatile networks by multi-

robot system. They developed a hybrid routing

protocol, i.e. proactive and reactive routing protocol,

demonstrated a high data transfer rate and showed

low recovery time in various cases. Their protocol

could cope with frequent network failures in not-so-

good network topologies. The authors claimed that

their protocol provided the best compromise

between latency and throughput for sparse highly

volatile networks. They, however, acknowledged

that routing protocols solve only part of problems in

multi-robot systems. They were aware of the need

for methodologies for maintaining efficient

connectivity of nodes while simultaneously

achieving task goals. We believe our humble study

can contribute this line of investigation. We are not

aware of any study of creating optimal route with

minimum number of relay robots.

As Voyles et al. have done, we have employed

Bluetooth ad hoc wireless communication called

scatternet (Cuomo et al., 2004). A scatternet is a

number of interconnected piconets that supports

communication between Bluetooth-equipped devices.

Figure 1 shows a scatternet that consists of three

piconets. A piconet is the type of connection that is

formed between two or more Bluetooth-equipped

devices. Since a piconet consists of one master node

and at most seven slave nodes, it can only handle at

most eight devices. Therefore a considerably large

scale ad hoc net must be formed by using scatternet.

Scatternets can be formed when a member of one

Self-OptimizingAlgorithmsforMobileAdHocNetworksbasedonMultipleMobileAgents

157

piconet (either the master or one of the slaves) elects

to participate as a slave in a second, separate piconet.

The device participating in both piconets can relay

data between members of both ad hoc networks.

Figure 1: Three piconets construct a scatternet.

One of the most closely related previous researches

is the “chain based path formation” of swarms of

robots conducted by Nouyan and Dorigo (Nouyan

and Dorigo, 2006). The concept of robot chains

stems from Goss and Deneubourg (Goss and

Deneubourg, 1992). A similar system was

implemented by Drogoul and Ferber (Drogoul and

Ferber, 1992). As these previous approaches, in the

system of Nouyan and Dorigo, every robot in a

chain emitted a signal indicating its position in the

chain, and they utilized the “cyclic directional

patterns” in order to give the chains directionality.

Unlike our proposing system, their purpose of

the research was investigating the capabilities of the

swarm robots that were self-organizing into chains

from random positions. They have found the impact

of the two parameters which determine the rate at

which a robot aggregates into, and disaggregates

from, a chain. They have also shown that their

system scales quite well with respect to the number

of robots. They, however, did not claim any

particular application for that chain forming, and

they stated that they were interested in studying

control algorithms that allow swarm of robots to

form arbitrary shapes instead of serializing.

Our purpose, on the other hand, is improving an

already established MANET connection with

arbitrary two points with scattered mobile robots

with minimum costs.

3 ALGORITHMS

The basic concept of optimizing the arrangement of

mobile robots that relay ad hoc communication is to

serialize the relay robots, and then to make

redundant robots leave from the relay line as shown

in Figure 2.

In order to make the relay robots move to form a

line, it is necessary to obtain the vector value of each

pair of adjacent robots. In order to accomplish this

we employ a mobile software agent to travel from

the source robot (robot A) to the destination robot

(robot E) as shown in Figure 3.

Figure 2: Optimal configuration.

Figure 3: The mobile agent is created at the source robot

and travels toward the destination while obtaining the

vector values.

Assume robot A is communicating with robot E by

using ad hoc wireless communication with several

relay robots. We also assume that both of them are

engaging some tasks at their current locations and

cannot move. Each time the mobile software agent

migrates one robot to another robot (one hop); it

checks the source robot through the camera on the

destination robot and obtains the vector value from

the destination to the source robot. Therefore, when

the mobile agent arrives at the final destination robot

(robot E), it has a sequence of vector values of all

the pairs of adjacent relay robots.

Upon arriving at the destination robot, the

software agent goes back the same route from the

destination robot to the source robot, and gives the

corresponding vector values to all the relay robots as

shown in Figure 4. When the mobile agent

distributes the vector value to each corresponding

relay robot, it adjusts the vector value so that it

points to the destination node robot. When the

mobile agent arrives at the source robot where that

agent was created, and the agent completes

distributing all the vector values it has collected

during the forward travel, its task is over and

vanishes.

Figure 5 shows the vector values given by the

mobile agent. Each letter represents the end robots

end robot

relay robot

connection

A A E E

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

158

and relay robots. Robot A is the source robot where

the mobile agent was created and E is the destination

robot. Since robot A and E cannot move, we want to

make other robots B, C, and D move to form a

straight line (optimal formation) to relay the ad hoc

connection.

Figure 4: The mobile agent distributes the vector values to

all the relay robots.

Figure 5: Each robot has its vector value.

3.1 Algorithm 1: Move All the

Participants

The moving algorithm makes all the participating

robot that are relaying the communication from the

source robot, i.e. robot A, to the destination, robot E,

move to form a straight line from A to E. The

algorithm consists of two phases. The first one is to

form a straight line, and then the second phase finds

redundant relay robots.

The idea is as follows. As shown in Figure 6, the

desired straight line is AE; therefore we want to

move the relay robot at point B to point B’. In order

to achieve this requirement, we need to obtain the

vector value ′





as follows:

′





















(1)

Figure 6: The first phase; a relay robot moves to an

internal dividing point.

Since 





and 





are known values, and m is the

number of hops from the source point of robot A and

n is the number of hops from the destination point of

robot E, it is straightforward to calculate the vector

value ′





When applying this algorithm to the relay robots,

all the relay robots move to the internal dividing

points on the line AE, and distances between

adjacent robots are shorten. Then some redundant

robot must be produced. Redundant robots means

two or more relay robots exist in a range of ad hoc

connection. When a relay robot recognizes it is

redundant itself, it tries to leave from the connection.

We describe the leaving algorithm in Section 3.3 in

detail. After successfully forming a straight line in

the first phase, the robots on the connection

sequence start to eliminate further redundancy in the

second phase as follows.

The second phase begins by dividing the relay

robots into roughly two groups, the left half and the

right half as shown in Figure 7.

Figure 7: The second phase; a robot in the left group

moves toward right-hand side and a robot in the right half

group moves toward left-hand side to find a redundant

robots.

The relay robots in the left half group move toward

the right hand side as far as they can maintain their

connection to the adjacent relay robots, and the relay

robots in the right half group move toward the left

hand side also as far as they can maintain their

connection, so that some relay robots can have new

and redundant connection.

When new connection is produced and a relay

robot becomes redundant, it tries to leaves from the

connection. We describe how a redundant relay

robot departs from the connection in section 3.3.

3.2 Algorithm 2: Move Minimum

Number of the Participants

Since moving all the participating relay robots are

rather inefficient, it is desirable if we can move

minimum number of robots to form the same

straight line. For this purpose, we extend the

algorithm 1 as follows. First, we number all the

relaying robots and if we have more than twice as

many robots as we need to construct a straight line

connection, we only move the even numbered robots.

If we do not have such enough robots, we choose the

least necessary number of robots from the source

E E

Self-OptimizingAlgorithmsforMobileAdHocNetworksbasedonMultipleMobileAgents

159

robot side, and move them.

Since we know the straight line distance from the

source robot to the destination robot, it is

straightforward to calculate the least number of

robots to maintain the connection.

3.3 Relay Robot’s Leaving from the

Connection

As described in the previous sections, when we have

succeeded in forming a straight line connection of

the relaying robot, we have some redundant robots.

In this section we describe how to find the redundant

relay robots, and how to make them leave from the

connection.

When the relay robots move based on the

syntheses of vector values described in the previous

section, some of them find new connections. Figure

8 shows the situation that moving relay robots (blue

ones) find a new connection, and one of them

becomes redundant.

Figure 8: A redundant robot leaves from the connection.

When a relay robot finds a new connection, it

notifies its finding to the neighbouring robots. If a

relay robot receives such notifications from both of

the adjacent robots, it recognizes it is the redundant

node of the connection (Figure 8-3). Then that relay

robot requests both of the adjacent robots the

permissions of leave (Figure 8-4), and if it receives

the acknowledgements from both of them (Figure 8-

5), it disconnects and leaves from the connection

(Figure 8-6).

When two adjacent relay robots find them

redundant and request for leave simultaneously, the

request-for-leave messages make collision each

other as shown in Figure 9. In such situation, the

two relay robots cancel their request for leave and

try again after randomly selected waiting time.

Figure 9: The collision of the request-for-leave.

Figure 10 shows another case of disconnection of relay.

In this case, the relay robot C wants to move upward to

straighten the connection A to D. But to do so, it must

leave the connection range with robot E. If it finds

connection to robot E is not active at that time, it cuts the

connection to E and moves outside of the connection

range of E, and otherwise it stays in the connection range

of E. In that case, it cannot move and stay at the current

position.

Figure 10: A relay robot disconnects to move to the

optimal location.

4 NUMERICAL EXPERIMENTS

In order to demonstrate the effectiveness of our

algorithms in a realistic environment, we have

implemented a simulator for ad hoc networks based

on multiple mobile robots, and conducted numerical

experiments. On the simulator, communication

scope, moving and rotating speed of robots, and time

lags required in agent migration and object

recognition are based on the data obtained from the

preliminary study using a herd of i-Robots Create

and Bluetooth scatternet. In the experiments, we set

the following conditions:

1. Robots are scattered in a 440×380 rectangular

field in the simulator.

2. The number of the robots is one hundred.

3. Each robot is represented as a circle that radius is

five.

4. The communication range of each robot is

seventy-five.

5. The distance that each robot can move in one

step is two.

6. The coordinates of the source and destination

robots are (10, 10) and (430, 370), respectively.

request for leave

uest for leave

find new connection

uest for leave

permission for leave

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

160

7. Their initial locations of other ninety-eight robots

are randomly decided without overlapping.

The Figure 11 shows the initial configuration.

The green lines indicate the established links of

robots via wireless ad hoc network. The red nodes

indicate the robots that are contributing to the

communication from the source node robot at the

upper left corner to the destination node robot at the

bottom right corner. Eleven robots are participating

in forming a connection from the source robot to the

destination robot. The blue nodes indicate robots

that are not participating in that particular

communication.

Figure 11: The initial configuration.

The Figure 12 shows the stable configuration after

216 steps in the simulation that employs the first

algorithm that moves all the participating robots.

We have observed that four robots out of eleven left

the connection. The report says the total distance

and total angles all the participating robot move and

rotate are 814 and 3990, respectively.

The Figure 13 shows the stable configuration

after 150 steps in the simulation that employs the

second algorithm that moves minimum number of

robots. The algorithm also produces connection

with seven robots, that is the optimal configuration

from the source robot to the destination robot, but it

moves only seven robots out of eleven. The total

distance and total angles that seven robots move and

rotate are 452 and 580, respectively.

From the observation above, it may appear that

the algorithm that moves minimum number of robots

is superior to the algorithm that moves all the

participating robots. But we have found that the

algorithm that moves minimum number of robots

has not always succeeded in reducing the number of

the relay robots minimal. The above example is the

ideal case. The Figure 14 shows the success rates of

the two algorithms. The two graphs show how

much percent could actually depart from the

connection successfully. The Figure 14a shows the

case of moving all the participating robots and

Figure 14b shows the case of moving the minimum

number of robots. The algorithm that moves all the

participants successfully remove all the redundant

robots (100%) sixty-three out of hundred patterns.

On the other hand, Figure 14b shows that the

algorithm that moves minimum number of robots

cannot achieve such successes. The algorithm failed

to remove entire redundant robots in most cases.

Figure 12: The stable configuration after moving all the

participating relay robots.

The reason why the second algorithm that moves the

minimum number of relay robots shows such low

success rate is frequent occurrences of deadlocks.

As we mentioned in the previous section, a relay

robot that have plural active connection often cannot

move. Even in one sequence of connections, we

have found frequent deadlocks. In contrast, the first

algorithm makes all the participating robots move

toward the same straight line, the robot rarely stack

in deadlocks.

The Figure 15 and 16 show the moving distances

and rotation degrees, respectively. The algorithm

that moves all the participants takes twice as long as

the algorithm that moves only minimal participants.

Also the former algorithm takes three times as much

degree as the latter algorithm. This phenomenon can

be easily understood, because the algorithm that

moves all the participants consists of two phases as

well as moves more number of robots.

In addition to the inefficiency, the algorithm that

moves all the participants has one big disadvantage.

That is the number of disconnections of network

links. We have found the algorithm that moves all

Self-OptimizingAlgorithmsforMobileAdHocNetworksbasedonMultipleMobileAgents

161

Figure 13: The stable configuration after moving the

minimum numbers of participating relay robots.

(a): The success rate of moving all the participants.

(b): The success rate of moving the minimum participants.

Figure 14: The success rates of the two algorithms.

the participants disconnect links about three times as

many as the algorithm that moves minimal

participants. Moreover, the algorithm that moves all

the participants tends to change the network

topologies and thus produces many disconnected

robots.

Therefore if we focus to provide optimal

connection only between certain two nodes, the first

algorithm that moves all the participants excels at

Figure 15: The moving distances.

Figure 16: The rotating degrees.

forming the optimal configuration. If we observe

wider scope, however, and find several ad hoc

connections request their own optimal formations

simultaneously, we may have another story. In such

cases, the side effects produced by moving relay

robots for one sequence of connections affect other

sequences of connections. The situation where

multiple ad hoc connections are active in parallel is

so complex that measuring the side effects are hard

to accomplish.

In our present study, we consider only one

sequence of connections, and have to conclude that

moving all the relay robots almost always provides

the optimal configuration, but that algorithm may

produces side effects that we yet to know how

harmful they are. We only know that the fewer the

moved relay robots, the less side effects occur.

As the future work we need to investigate the

situation where multiple connections of ad hoc

wireless network exist simultaneously. The situation

should not be difficult to handle; simply we need to

add one mobile agent for each connection. Then the

autonomous mobile software agent should all the

jobs. We believe our model is quite scalable. We

only need to polish the first algorithm so that the

entire moving cost is minimal, or to polish the

second algorithm so that the success rate is high.

all move

minimum

move

minimum

move

distance per robot total distance

degrees per robot total degrees

times

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

162

5 CONCLUSIONS AND FUTURE

WORK

We have presented two algorithms that form optimal

configuration of ad hoc networks with multiple

mobile robots. One algorithm moves all the

participants and succeeds in configuring the optimal

connection (minimal number of relay robots) more

than sixty percent. But this algorithm naturally takes

more time to reach stable configuration and moves

more robots, and thus consumes more energy. The

other algorithm moves only minimum participants

and often fails to produce optimal connection. It

often fails to eliminate redundant robots too. But

this algorithm is naturally more efficient. For

connecting certain two nodes, the algorithm that

moves all the participants provides better result.

However, this algorithm changes the network

topologies and thus produces more disconnected

robots. When we consider the network topologies

changes a lot in multiple robot environments, and

such environments need to connect arbitrary pairs of

nodes, this side effect may cause serious problem.

Therefore we need to investigate the algorithm that

moves minimum participants and improve the

success rate of that algorithm.

An additional problem may occur in the cases of

applications of both algorithms, due to the constraint

of piconet. Since Bluetooth allows a master can

have only seven slaves, if a master already has the

maximum number of slaves, it cannot connect to a

new node even though it finds a new node as shown

in Figure 17. In order to establish a new connection,

it must cut one of the existing connections.

Selecting the most promising relay robots is a big

problem worth to investigate. We plan to pursue

this direction too.

Figure 17: Too many slaves.

REFERENCES

Binder, W., Hulaas, J. G., and Villaz, A., 2001. Portable

resource control in the j-seal2 mobile agent system. In

Proc. of the fifth international conference on

Autonomous agents, AGENTS ’01, 222–223.

Correll, N., Bachrach, J., Vickery, D., Rus, D., 2009. Ad-

hoc Wireless Network Coverage with Networked

Robots that Cannot Localize, In IEEE International

Conference on Robotics and Automation.

Cuomo, F., Melodia, T., Akyildiz, I. F., 2004. Distributed

self-healing and variable topology optimization

algorithms for QoS provisioning in scatternets, IEEE

Journal on Selected Areas in Communications, 22(7),

1220-1236.

Drogoul, A., Ferber, J., 1992. From Tom Thumb to the

dockers: Some experiments with foraging robots. In

From Animals to Animats 2. Proc. of the 2nd Int. Conf.

on Simulation of Adaptive Behavior (SAB92), MIT

Press, Cambridge, MA, 451–459.

Goss, S., Deneubourg, J.L., 1992. Harvesting by a group

of robots. In: Proc. of the 1

European Conf. on

Artificial Life, MIT Press, Cambridge, MA, 195–204.

Heo, N., Varshney, P. K., 2003. A Distributed Self

Spreading Algorithm for Mobile Wireless Sensor

Networks, In IEEE Wireless Communications and

Networking Conference.

Kambayashi, Y. and Takimoto, M., 2005. Higher-Order

Mobile Agents for Controlling Intelligent Robots.

International Journal of Intelligent Information

Technologies, 1(2), 28–42.

Kambayashi, Y., Yamachi, H., Takimoto, M., 2012.

Feasibility Studies of the Intelligent Cart System.

Communications in Information Science and

Management Engineering, 2(6), 1-8.

Nguyen, C. Q., Leong, B., Matson, E. T., Smith, A., ．

Wachs, J. P., 2010. AWARE: Autonomous Wireless

Agent Robotic Exchange, In ICIRA 2010, 3

International Conference on Intelligent Robotics and

Application, LNAI 6276, 198-208.

Poduri, S., Sukhatme, G. S., 2004. Constrained Coverage

for Mobile Sensor Networks, In IEEE International

Conference on Robotics and Automation, 165-172.

Parker, L. E., 2008. Distributed intelligence: overview of

the field and its application in multi-robot systems,

Journal of Physical Agents, 2(1), 5-14.

Nouyan, S., Dorigo, M., 2006. Chain Based Path

Formation in Swarms of Robots, In Ant Colony

Optimization and Swarm Intelligence, LNCS 4150,

120-131.

Shintani, M., Lee, S., Takimoto, M., Kambayashi, Y.,

2011. A Serialization Algorithm for Mobile Robots

Using Mobile Agents with Distributed Ant Colony

Clustering. In Knowledge-based and Intelligent

Information and Engineering Systems, (KES 2011).

LNAI 6881, 260-270.

Voyles, R. M., Bae, J., Larson, A. C., Ayad, M. A., 2009.

Wireless video sensor networks for sparse, resource

constrained, multi-robot teams, Intelligent Service

Robotics, 2(4), 235-246.

Self-OptimizingAlgorithmsforMobileAdHocNetworksbasedonMultipleMobileAgents

163