Multi-agent Approach for Return Route Support System Simulation

Shouhei Taga

, Tomofumi Matsuzawa

, Munehiro Takimoto

and Yasushi Kambayashi

Department of Computer & Information Engineering, Nippon Institute of Technology,

Saitama, Japan. 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: Multi Agent, Mobile Agent, Ant Colony Optimization, Mobile Ad Hoc Network, Contingency Plan, Risk

Management.

Abstract: We propose a system that supports stranded commuters caused by a large-scale disaster. When a large-scale

disaster breaks out, buildings may collapse and roads may be damaged and the public transportation systems

would be paralyzed. Thus, people working in the city center have to walk back home on foot. The problem is

that when those people start walking, the situation along the routes for returning home may be different from

that of the pre-disaster. Not only may it be the first time for most of them to walk home, but also the return

route may be extremely complex due to many detours. They have to look for alternative routes whenever

bridges collapse and fires break out. Making situation become worse, modern people intensively use

navigation systems, those systems may be unavailable due to the paralyzed Internet. A large scale disaster

may destroy base stations of wireless phones, and even if it does not completely destroy them, extreme

congestion may paralyze the communication infrastructure so that not only net-surfing using smartphone, but

also collecting information by e-mail may become impossible. To deal with such situations, we are designing

a system that provides those unfortunate pedestrians appropriate return routes to their homes without

depending on the communication infrastructures. Instead, our proposed system only depends on smartphones

of those pedestrians and constructs mobile ad hoc networks (MANET) to collect and disperse useful

information. We employ multiple mobile agents extensively for information collection and dispersion. In

order to demonstrate the feasibility of our system, we have constructed a preliminary prototype of the

simulation system and have conducted numerical experiments.

1 INTRODUCTION

When the Great East Japan Earthquake occurred in

2011, we observed many cracks and liquefaction

phenomena on the ground as well as communication

infrastructure broke down even far from the directly

affected areas, i.e. Tokyo. Tokyo is more than 300km

south of Sendai, a major city close to the seismic

centre, even though the centre of the shock was the

sea bed. Those phenomena made public

transportations paralyzed. On top of the unfortunate

events, the communication network became

unreliable due to the collapse of base stations and

congestion of communication networks. The situation

made people who started walking home be unable to

collect information such as the state of the return

routes using the Internet. A large number of stranded

commuters had to return home without any

information. In a modern society, too many people

rely on the communication network environment and

depend on information gathering from the Internet. It

is not difficult to imagine that how hard and

dangerous for those modern people who heavily

depend on the Internet, e.g. the navigation systems, to

suddenly start walking without any Internet

assistances. They may have to face possible

secondary disasters.

Actually, there are several services that provide

return routes for people walking back home. For

example, Mapple ON provides an application for

returning home support map when a disaster occurs

(Mapple ON Co., Ltd., 2013). This application guides

the user to support facilities such as first aid stations

as well as provides return route to homes by using

GPS information. This application utilizes the

information such as road networks. This system,

however, uses the information before a disaster

occurs. Therefore when the disaster makes some

roads impassable, the recommended routes by using

Taga, S., Matsuzawa, T., Takimoto, M. and Kambayashi, Y.

Multi-agent Approach for Return Route Support System Simulation.

DOI: 10.5220/0005819602690274

In Proceedings of the 8th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2016) - Volume 1, pages 269-274

ISBN: 978-989-758-172-4

269

the map information become useless. We need a

guiding system that uses the real-time information

without Internet support.

In this paper, we propose a return route support

system that uses real-time information collected by

not using the Internet. That is a system temporarily

constructs a MANET environment that connects the

smartphones for communication means and makes

them share useful information for returning home.

The system collects as well as shares the latest

information in the MANET using mobile agents for

sharing it. By using the collected information, the

system derives a certain return route for each person.

Henceforth, we call this disaster return home route

support system simply "the system", and the stranded

commuters to take advantage of this system "the

users".

When the user finds an impassable point of a road,

he or she is expected to dispatch a mobile agent to

spread the information to share it with other users. In

order to collect information for the return route, the

user dispatches another mobile agent to smartphones

in the destination direction to look for useful

information. Thus, the user can obtain the optimal

return route in light of the current road conditions.

It is essential to use map information for deriving

the return route. In the post disaster environment,

however, it is difficult to acquire the map information

from the Internet services. In this system, the user is

expected to register the map data that covers the home

and office before the disaster occurs. This way, the

system can deal with the map information in a disaster.

The structure of the balance of this paper is as

follows. The second section describes the background

and discusses the related works. The third section

describes the multi-agent system that is the basis of

our proposed system. In order to demonstrate the

feasibility of our system, we have constructed a

prototype of the simulator for this system and have

conducted numerical experiments. We report the

results and discuss the future works in the fourth and

fifth sections. Finally we conclude our discussion in

the sixth section.

2 BACKGROUND

As we start our study, there are two major studies

closely related to our system. One is the return route

generation methods proposed by (Abeta et al., 2007).

In order to assist people who return home in a disaster,

their system collects the trajectories of stranded

commuters using an ad hoc network, and proposes

return routes using the trajectories. The users can then

obtain the recommended moving directions through

their mobile phones from the terminals associated to

nearby intersections. By numerical experiments, they

have showed that they have succeeded to derive good

return routes.

The other one is Alejandro Aviles’s ERAM

(Evacuation Routing using Ant Colony Optimization

over Mobile Ad hoc Networks) (Alejandro et al.,

2013). In his study, in order to derive the evacuation

routes, he employs a multi-agent system and the ACO

(Ant Colony Optimization) in an assumed MANET

environment. In his simulation system, when an

evacuee successfully exits the building (the system

assumes to be used indoor), a mobile agent called

goal agent (GA) is generated from the smartphone the

evacuee has. The generated GA moves to the other

smartphone and records the number of movements.

The number of movements acts as the pheromone in

the ACO. The fewer the movements, the better

trajectory the system evaluates and route seeking

agents are probabilistically attracted. Routing Agent

(RA) is the route seeking mobile agent that is also

generated from the evacuee’s smartphone. RA moves

to the other smartphones and records the number of

movements of the GA. Then the RA recommends the

optimal evacuation route as the trajectory of the GA

that has the least number of movements.

Asakura et al. have investigated the way

information is exchanged between evacuees in a

disaster. They have constructed a map information

sharing system (Asakura et al., 2011). In their system,

some evacuees record road information while moving

to shelters, and then exchange the information they

have collected using a MANET. Since the situation

can change from time to time in a disaster area, they

have proposed an ACO-based extension to allow

evacuees to respond to such changes (Asakura et al.,

2013).

In our proposed system, we have also introduced

MANETs and multi-agent system. A MANET is a

communication network formed by only mobile

terminals. The difference between existing

communication systems and MANETs is that

MANETs do not require the base stations. When a

large-scale disaster occurs, it may become difficult to

conduct communication using the base stations. In

addition, communication failure may also occur due

to congestion. Therefore, MANET is suitable for

constructing a temporary communication means

without using the base stations. We have taken

advantages of Wi-Fi for constructing communication

links between smartphones. One of the disadvantages

of MANETs is the fact that links are disconnected

frequently because smartphones that constitute the

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

270

network are literary moving. We can mitigate this

problem by using mobile agents. Mobile agents are

software agents that move between devices. Mobile

agents provide a means of communication where

connections are intermittent, because connections are

only needed when they move. Thus, it is not

necessary to keep connections during the entire

communication processes.

In addition, we have employed the concept of the

pheromone in the ant colony optimization (ACO)

(Dorigo et al., 1996). ACO is an algorithm that

mimics the foraging behaviour of ants. When ants

move back and forth between food and nest, they

deposit a chemical substance called pheromone.

Pheromone evaporates over time. The ants following

the pheromone trace to reach the food also deposit

pheromone. As a result, the route that connects from

the nest to the food in the shortest path receives more

pheromone before its evaporation than other longer

routes, and thus the most convenient route is

established.

Even though we have proposed the base algorithm

of this system in the previous paper (Yatsuyanagi et

al., 2014), the simulator based on the algorithm was

too crude, and the study failed to produce reasonable

return routes. In this proposed system, we succeeded

to derive a reasonable path by using Dijkstra’s

algorithm (Dijkstra, 1959) in which the intersections

are treated as the nodes.

3 AGENT SYSTEM

In this system, we adopt an agent system comprising

multiple mobile software agents. In the agent system,

there are two types of agents. One is mobile agents

that move among the users’ smartphones: they are the

routing agent (RA) and the information diffusion

agent (IDA). The other is static agents that remain on

the users’ smartphones: they are the information

agent (IA) and the node management agent (NA).

The roles of the mobile agents are collecting

information while moving among the smartphones in

the MANET as well as diffusing the newly

discovered information such as impassable points and

first aid stations. The roles of static agents are storing

and organizing information that is conveyed by the

mobile agents, and deriving the routes to the

destinations based on the collected information, as

well as providing user interface for receiving input

from the users. Figure 1 shows the relations between

these agents. We describe each agent in detail in the

following subsections.

Figure 1: Information agent (IA) generates and dispatches

RA and IDA for collecting and diffusing route information.

3.1 Information Agent (IA)

The roles of IA are providing the user interface,

generating mobile agents, and deriving the return

routes.

First, we describe the user interface. IA displays

the derived route on the map with useful information

such as impassable points and first aid stations. When

the user finds an impassable point or a first aid station,

he or she presses the corresponding diffusion button

on the screen so that IA creates an IDA to diffuse the

newly discovered information. IA updates the

information on the map as RA returns and gives IA

new information.

Second, we describe the generation of mobile

agents. IA periodically generates RAs for collecting

information. When the IA generates an RA, it

requests NA and receives the destination location and

the ID of the smartphone of the user. After passing

the information to the RA, the IA dispatches the RA

toward a smartphone in the direction of the user’s

destination. The RA then moves from a smartphone

to another smartphone, and comes back with

collected information. When RA comes back, IA

receives the information from the RA and passes it to

NA. Then it generates another RA after a certain time

elapses. If RA has not come back after a certain time

period, IA regards the RA as lost, and generates a new

RA. When RA generated by another smartphone

arrives, IA requests NA, receives information and

passes it to the RA with its own ID. IA controls all the

information transfers.

When the user presses the diffusion button, IA

generates an IDA. When IA generates an IDA, it

passes type information and location information to

Multi-agent Approach for Return Route Support System Simulation

271

indicate what and where the user finds along with the

ID of the smartphone to the IDA, and dispatches it to

one of the smartphones nearby. Then, the IDA is

gradually diffusing into the set of other smartphones.

When an IDA generated by another smartphone

arrives, IA receives the information and passes it to

NA, and gives its ID to the IDA.

Third, we describe the route derivation. IA derives

the route to the destination by using the collected

information. While the default destination is set to the

home, the system switches the destination to the rest

area after certain time has elapsed since system

startup, in order to give the user a break. The user can

choose whether the switching occurs and duration

time for switching. IA derives the return route by

Dijkstra algorithm, and then displays it in the map on

the interface screen. If IA receives information of an

impassable point on the current route, it re-derives a

new route that avoids that point. IA also re-derives the

route when destination is altered.

3.2 Node Management Agent (NA)

NA keeps the collected information and passes the

information to IA on request. The stored information

includes GPS information representing the current

location, the current destination, the impassable

locations, the first aid stations and the location of the

rest areas along with the time they are stored and the

IDs of the IDAs that convey the information. When

redundant information is delivered, NA deletes the

old information by referring to the time information.

Also, after a certain time period from the store, the

old information is deleted.

3.3 Routing Agent (RA)

RA is the harbinger of the user. It conducts

reconnaissance of the direction toward the destination

and collects the information.

RA is generated by IA. Then RA moves to one of

the smartphones that is in the user's moving direction.

Upon arrival at such a smartphone, RA requests the

IA on the smartphone to provide information that it

has. Upon receiving the information, RA moves again

to another smartphone. After repeating this moving

operation some predetermined number of times, it

tries to return to the original smartphone. Since each

RA has the history record of the visited smartphones

as a sequence of IDs, RA can move back to its origin

by tracing the recorded smartphones. If a link of the

smartphones is lost, it waits to between for the link to

recover for a while. After waiting for a predetermined

period, and if the link is not recovered, the waiting

RA self-destructs. When RA succeeds to return to the

original smartphone, it passes the collected

information, and disappears. Figure 2 shows how an

RA moves.

Figure 2: RA’s movements.

3.4 Information Diffusion Agent (IDA)

IDA diffuses useful information such as impassable

points on roads and bridges, resting places and the

first aid stations the user finds. IDAs collectively

implement the information sharing among the users.

When the user presses one of the diffusion buttons

on the interface screen, IA produces an IDA. The IDA

has the information about what the user finds the

location and the time. Then, the IDA clones itself to

number of smartphones that are linked to the

originating smartphone. One copy moves each of the

neighbour smartphones. IDA has a unique ID and all

the IDs of the duplicated IDAs are the same. Upon

arrival, IDA passes its information and its ID to the

IA of the smartphone it arrived to and it clones itself

again to all connected smartphones. If its ID is

already stored, the IDA self-destructs without doing

anything. By doing so, IDA prevents dispersing

redundant information. Also, each IDA disappears

after a certain number of movements.

IDAs collectively implement the pheromone in

ACO. IDA determines the value of its information

according to the distance between the discovery point

and the current position, the number of movements it

performed, the elapsed time from it was generated. In

other words, IDA reduces the value of its information

as the number of movements, the elapsed time, the

distance between the discovery point and current

position increases. When the information value gets

to be below a certain threshold, the IDA gets to be

useless and self-destructs.

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

272

4 PRELIMINARY EVALUATION

To demonstrate the feasibility of our proposed multi-

agent system, we have constructed a simulator and

have conducted preliminary experiments on this

simulator.

The scenario of the simulator is as follows; after a

disaster occurs, the users move through a city, in

which a few impassable points exist, to the

destination using this system. Figure 3 shows a screen

shot of the simulator that displays the users (red dots),

an impassable point (purple area), a safety zone

(green region), and the links between the smartphones

(connecting lines between the users).

Figure 3: Simulator screen.

The number of users can be specified. Each user’s

smartphone has certain residual quantity of the

battery power remaining that is randomly set. When

the battery is dead, the user has to stop moving. Each

smartphone has certain communication range, and

when a user with a smartphone comes into

communication range, a link is automatically

established. When a user finds an impassable point

(comes within a certain view range), he or she

diffuses that information with GPS coordinates to

other users by dispatching IDAs. The users, who

receive that information by an IDA, as well as the

discoverer, try to find alternative routes to detour the

impassable point.

We have set two cases. In one case, all the

evacuees have smartphones equipped the system, and

in the other case, no one uses the system. We have

performed the simulation experiments ten times each

and compared the averaged consuming rounds for all

the users’ arrivals to the destinations. One round

means the duration time all the users execute one step.

In the experiments, we set the number of users 300,

the field size 800 × 625 pixels, and 2 pixels per round

as the amount of movement of a user.

Table 1 shows the results of the experiments.

Since the users who use the system are aware of the

impassable points on the route in advance, those who

use the system can arrive earlier to their destination

than those who do not use the system, as expected.

Table 1: Result.

Num. of rounds

Evacuation with the system 583

Evacuation without the system 683

5 DISCUSSIONS

Even the preliminary experiment shows the system is

feasible and effective; we have to investigate quite a

few problems in order to make the simulation system

be realistic.

A smartphone has limited resources. In the present

design, if the users observe many impassable points,

they dispatch many IDAs. Thus each smartphone in

the system may be overloaded. We have to consider

how to save the battery consumption due to the heavy

computational load of each smartphone. In order to

mitigate this problem, we can restrict the inflow of

IDAs. In order to set the threshold to restrict the

inflow, IA should sense the environment and set the

most appropriate value. However, that method may

have IA overloaded by heavy computations.

Therefore, it may be better to set a threshold in

advance. It is difficult to evaluate the appropriate

values for various environments. It is the next step we

have to take by implementing much more refined

simulation of how the respective methods affect the

system’s performance.

In the present system, RA is designed to go back

to the smartphone where it was created by tracing

back its movement history record. Therefore, if RA

cannot find one of the smartphones in the record, it

cannot return to the origin. There are many factors

that cause the trace of the smartphone to be lost: the

user of the smartphone may go somewhere, the user

may terminate using the system, and the battery of the

smartphone may be dead. In the simulator, the user

does not terminate the system until either he or she

succeeds in reaching the safe area, or the battery of

the smartphone dead. However, in reality, when the

battery capacity is running low, the user usually

switches off the smartphone to save the battery.

Therefore the possibility the RA loses the trace may

be higher in reality than in the simulator.

Multi-agent Approach for Return Route Support System Simulation

273

In order to solve this problem, we are planning to

provide alternative return routes for the RA that are

different from the ones in the movement record. For

example, when a RA arrives at the destination

smartphone, then the RA acquires information of

other smartphones connected to that smartphone. If

the RA cannot find the next smartphone in the

movement record, it moves to another smartphone

that is connected to the smartphone that may be

connected to the scheduled destination. By doing so,

we can increase the possibility of each RA to reach to

the original smartphone even if a RA cannot find the

smartphone in the movement record. However, this

technique imposes more burdens on RAs and may

causes performance degradation in the system.

In a wider context, Stranders et al. proposes

decentralized coordination algorithms for multiple

sensors (Stranders et al., 2009), and Zambonelli’s

SAPERE project is pursuing a pervasive services in

context-aware systems (Anzengruber et al., 2013;

Montagna et al., 2013). Their approaches are similar

to our proposal approach even though none of them

considers using MANET and mobile agents. We plan

to re-design our system after collecting data from the

simulation, and we will then integrate the concepts of

such competitive systems.

6 CONCLUSIONS

In this paper, we propose a return route support

system for stranded commuters going back home.

The system consists of smartphones connected

through MANETs. The users of smartphones can

exchange and share useful information by using

multiple mobile agents. In order to examine the

feasibility, we have constructed a prototype of the

simulator and have conducted a preliminary

experiment. As we have expected, the users who use

the system can arrive to the safe area more quickly

than the users who do not use the system. As a future

direction, we are refining the simulator to investigate

the problems discussed in the fifth section. For this

purpose, it is necessary to increase the capabilities of

the simulator. In particular, we are implementing a

new simulator where the system perceives the traffic

of people and adjust its behavior dynamically.

ACKNOWLEDGEMENTS

This work is supported in part by Japan Society for

Promotion of Science (JSPS), with the basic research

program (C) (No. 25330089 and 26350456), Grant-

in-Aid for Scientific Research.

REFERENCES

Abeta, T., Suzuki, Y., Kawagoe, K., 2007. Route generation

methods for supporting return route in disaster.

DEWS2007, DEWS2007 D7-3, pp.1-8.

Anzengruber, B., Castelli, G., Rosi, A., Ferscha, A.,

Zambonelli, F., 2013. Social feedback in display

ecosystems, IEEE International Conference on Systems,

Man, and Cybernetics, pp.2893-2898.

Asakura, K. Chiba, T. Watanabe, T., 2011. A map

information sharing system among refugees in disaster

areas, on the basis of ad-hoc networks, The 3

International Conference on Intelligent Decision

Technologies, pp.367-376.

Asakura, K. Chiba, T. Watanabe, T., 2013. Construction of

navigational maps for evacuees in disaster areas based

on ant colony systems, International Journal of

Knowledge and Web Intelligence. Vol. 4, Issue 4,

pp.300-313.

Aviles, A. M., Takimoto, M., Kambayashi, Y., 2014.

Distributed evacuation route planning using mobile

agents. Transaction on Computational Collective

Intelligence XVII, LNCS 8790, pp.128-144.

Dijkstra, E. W. 1959. A note on two problems in connexion

with graphs. Numerische Mathematik, 1, S. pp.269-271.

Dorigo, M., Maniezzo, V., Colorni, A., 1996. Ant system:

Optimization by a colony of cooperating agents, IEEE

Transaction on System, Vol. 26, No. 1, pp.29-41.

Mapple ON Co., Ltd., 2013. Disaster return home support

map. http://mapple-on.jp/productskitaku2013-14/.

Montagna, S., Viroli, M., Fernandez-Marques, J. L., Di

Marzo, G., Zambonelli, F., 2013. Injecting self-

organization into pervasive service ecosystems. Mobile

Networks and Applications, vol. 18, no. 3, pp.398-412.

Stranders, R., Farinelli, A., Rogers, A., Jennings, N., 2009.

Decentralised coordination of mobile sensors using the

max-sum algorithm. 21st International Joint

Conference on Artificial Intelligence, pp.299-304.

Yatsuyanagi, N., Takimoto, M., Kambayashi, Y., 2014.

Design of a return route support system using multiple

mobile agents. SICE Annual Conference, pp.1571-1576.

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

274