A DESCRIPTION METHOD FOR MULTI-AGENT

SIMULATION MODEL UTILIZING TYPICAL ACTION

PATTERNS OF AGENTS

Taiki Enomoto, Gou Hatakeyama, Masanori Akiyoshi and Norihisa Komoda

Osaka University, Yamadaoka 2-1, Suita, Osaka 565-0871, Japan

Keywords:

Multi agent simulation, Description of events, Typical action patterns, Event ﬂow.

Abstract:

Recently, there are various proposals on tool for multi-agent simulation. However, in such simulation tools,

analysts who do not have programming skill spend a lot of time to develop programs because notation of sim-

ulation models is not deﬁned sufﬁciently and programming language is varied on tools. To solve this problem,

a programming environment that deﬁnes the notation of simulation model based on graph representation is

proposed. However, in this environment, we still need to write programs about a ﬂow of event and contents

of agents’ action and effect. So, we propose a description method for multi-agent simulation model utilizing

typical action patterns of agents. In this method, users write about designs of contents of event based on typical

action patterns which are “interrogative (4W1H) and verbs”, and designs of a ﬂow of event. In this paper, we

executed experiments that compare time needed for examinees to generate programs by a conventional method

and our programming environment. Experimental result shows the time to generate programs by utilizing our

programming environment less than that by utilizing a conventional one.

1 INTRODUCTION

Recently, multi-agent simulation (MAS) is expected

to be effective for simulation of complex system

such as biology, ecosystem, social system and eco-

nomics(MacNealy, 1999)(Axelrod, 1995). MAS is a

method of simulation that includes some autonomous

agents which act in spontaneous manners like a hu-

man and make interactions each other. The purpose

of MAS is to analyse phenomenon generated by in-

teractions of agents. Agents make action based on

rules and affect other agents.

MAS system consists of these local rules. There

have been some studies of MAS, and various MAS

tools have been developed (Swarm(Minar et al.,

1996), Repast(North et al., 2005), artisoc(kke)) and

used for many simulationDHowever,analysts spend a

lot of time to describe a simulation model of the prob-

lem, because there is no way of understandable nota-

tion of simulation models based on feature of MAS.

Moreover, when analysts write the program codes of

simulation, they have to use speciﬁc programming

language of MAS tool which makes them spend a lot

of time to acquire programming skills.

To solve the above-mentioned problems, a pro-

gramming environment that deﬁnes the notation of

simulation model based on graph representation is

proposed(Hatakeyama et al., 2007). However, in this

environment, we still need to write program codes

about a ﬂow of event and contents of agents’ action

and effect.

In this paper, we propose a description method

for multi-agent simulation model utilizing typical ac-

tion patterns of agents. In this method, users write

about contents of event based on typical action pat-

terns, which are “interrogative (4W1H) and verbs”,

and designs of a ﬂow of an event.

2 A MULTI-AGENT SIMULATION

PROGRAMMING

ENVIRONMENT

2.1 Outline of Programming

Environment

Analysts who run simulation roughly sketch interac-

tions of objects in simulation, and they design a simu-

lation model by embody their understanding. Next,

316

Enomoto T., Hatakeyama G., Akiyoshi M. and Komoda N. (2008).

A DESCRIPTION METHOD FOR MULTI-AGENT SIMULATION MODEL UTILIZING TYPICAL ACTION PATTERNS OF AGENTS.

In Proceedings of the Third International Conference on Software and Data Technologies - ISDM/ABF, pages 316-319

DOI: 10.5220/0001900303160319

 SciTePress

Figure 1: Outline of programming environment.

they write program codes based on the simulation

model and run simulation with MAS tools. We can

use several MAS tools as execution platforms, but

these tools do not support to describe a simulation

model and develop programs. To solve these prob-

lems, a programmingenvironment that deﬁnes the no-

tation of simulation model based on graph represen-

tation is proposed(Hatakeyama et al., 2007).

Figure 1 shows overall conﬁguration of our simu-

lation programming environment. First, analysts de-

scribe a simulation model of target problem by utiliz-

ing the editor for MAS model. Secondly, this sim-

ulation model is transformed into XML (eXtensible

Markup Language) to represent information of the

simulation model. This XML data is interpreted and

transformed into a simulation program code by the

program generater. We use “artisoc” as MAS execu-

tion platform. With MAS execution platform, we run

the generated simulation program code and get simu-

lation results.

2.2 MAS Model Editor

In our simulation programming environment,we need

some elements to deﬁne the model of a target prob-

lem. Figure 2 shows these elements and a descrip-

tion of model based on these elements. In this ex-

ample, the relationship between shops and consumers

is shown in the upper part of Figure 2. First, we

have to describe an agent itself and interaction of each

“agent” such as consumers and shops. An agent has

“attributes” which represent an agent’s state. Also, an

agent has “action” which inﬂuences other agents be-

havior or attributes. We need these 3 elements to de-

scribe a simulation model. Then the simulation model

based on graph representation is depicted in the lower

part of Figure 2. An agent’s attribute is represented

as “variable node”, an agent’s action is represented as

“event” and an effect is represented as “arc”.

An agent’s attributes, action and effect are repre-

sented as a node, an event and an arc in MAS model

respectively. However, contents of action and transi-

tions of event are not represented well. So, we pro-

pose a description method for multi-agent simulation

model utilizing typical action patterns of agents to

support users to describe such events.

Figure 2: A description method of model.

3 EVENT DESCRIPTION

3.1 Outline of Event Description

Contents of events that agents behave independently

are able to be described by utilizing “when” (when a

condition is satisﬁed), “who” (an agent), “how” (fol-

lowing the rule), “what” (object) and “verb” (do).

In this paper, we deﬁne “interrogative (4W1H) and

verbs” as typical action patterns of agents and pro-

pose a description method for multi-agent simulation

model utilizing typical action patterns of agents.

In addition to use of the interrogative (4W1H) and

verbs for expressing the contents of agent events, we

introduce “ﬂow diagram of event transitions” for ex-

pressing interactions between agents.

3.2 Design of Event Contents

Figure 3 shows outline of design of event contents.

We prepare pull-down choice lists to design event

contents about “interrogative (4W1H) and verbs”. In

case of designing event that is selection of shop and

if we want to design such as “if distance between a

customer and a shop is within a certain value that we

decide in advance, the customer selects the shop.”, we

describe a model such as “When => Always”, “Who

=> customer”, “How => Min (Distance (customer,

shop))”, “What => shop” and “Verb => Select One”.

A DESCRIPTION METHOD FOR MULTI-AGENT SIMULATION MODEL UTILIZING TYPICAL ACTION

PATTERNS OF AGENTS

317

Figure 3: Design of event contents.

Figure 4: Example of a ﬂow of an event.

There is a problem that it is difﬁcult to prepare

all verbs which are used whole entire world to design

event contents. So, in this paper, we deﬁned vocabu-

lary of verbs by putting the verbs that we use everyday

together through the lens of the agent world.

3.3 Design of a Flow of an Event

Figure 4 shows a ﬂow of an event. The transition of

an event is described about each agent in the ﬂow of

an event. The condition of transition is described in

each arc by referring to the vocabulary that is stated

in section 3.2. Also, timing of ﬁring the event and the

procedures that the event occurs are described in the

ﬂow of the event. In the Figure 4, the state “wait” rep-

resents that one agent waits for other agents to com-

plete an event that it is necessary to be completed for

one agent to transit next event.

4 AUTOMATIC GENERATION OF

PROGRAMMING CODES

MAS model that is described by utilizing a model de-

scription method based on graph representation are

represented in a form of XML. Event contents that

are included in the MAS model and information of

transition of events are represented in the same form.

The XML ﬁle is input into the program generater and

Figure 5: Outline of automatic generation of programming

codes.

converted to program codes. Figure 5 shows the out-

line of automatic generation of programming codes.

Also, we use “artisoc” as an execution platform in

this paper. In “artisoc”, description of agents is made

up of “declaration of value that agents have”, “agents

rules that are executed per unit time” and “deﬁnition

of function by users”. In Figure 5, MAS model corre-

sponds with declaration of value, design of event ﬂow

corresponds with agents rule and design of event con-

tents corresponds with deﬁnition of function.

5 EVALUATION

We evaluate our description method by executing ex-

periment on “shopping mall model”.

5.1 Generation of a Shopping Mall

Model

Figure 6 shows outline of a shopping mall.

• Customers arrive at the west gate or the south gate

by some probability.

• Customers move in the shopping mall freely.

• Shop B passes coupons at the west gate and cus-

tomers receive them at a constant rate.

• Customers go into the shop if distance between

customer and a shop is within a constant value.

• Shops do not allow customers to go into the shop

when each store capacity is not sufﬁcient. And

ICSOFT 2008 - International Conference on Software and Data Technologies

318

customers ﬂounce in the shopping mall freely

again.

• If customers in shop A have a coupon, they leave

the shop and do away with their coupon and they

ﬂounce in the shopping mall freely again. In other

cases, customers buy goods.

• When customers ﬁnish buying goods, they leave

the shopping mall

• The shops count the number of their customers

Figure 6: Outline of shopping mall.

5.2 Generation of Model by Examinees

5.2.1 Precondition of Experiment

First, examinees receive an explanation about the

shopping mall model that is described in section 5.1.

Secondly, they generate the shopping mall model by

utilizing our event description method. Finally, we

compare required time, which examinees generate a

model by utilizing our approach and by writing raw

program codes for “artisoc”.

Two examinees have programming experience,

however they inexperienced in MAS or “artisoc”. So,

they learn how to use “artisoc” by reading its manual.

5.2.2 Result of Experiment

Table 1 shows the result of experiment.

Table 1: Result of experiment.

In experimental result, it is clariﬁed that time,

which is required to generate a simulation model by

utilizing “artisoc”, is reduced by utilizing our event

description method. Additionally, our event descrip-

tion method is easy to use for not only person of ex-

perience but also amateur programmers of MAS.

However, simulation sometimes does not go well,

because users draw the model in wrong manners. For

example, in case of “leave shop event”, users have

to describe processing to reduce the number of cus-

tomers. However, because of a perceived notion that

the number of customers reduces naturally, users fail

to describe this procedure explicitly. So, this simula-

tion does not go well. We need to consider a method

to bridge the gaps between users’ perceived notions

and codes.

6 CONCLUSIONS

In this paper, we proposed a description method for

multi-agent simulation model utilizing typical action

patterns of agents in a programming environment that

deﬁnes the notation of simulation model based on

graph representation.

In this paper, we executed experiment and com-

pared required time, which examinees generate a

model by utilizing our approach and by “artisoc”. As

a result, it is clariﬁed that time is reduced by utilizing

our event description method. Additionally, an event

description method that we propose is easy to use for

not only person of experience but also amateur pro-

grammers of MAS.

REFERENCES

Axelrod, R. (1995). The convergence and stability of cul-

tures : Localconvergence and global plarization. Tech-

nical report, Working Paper 95-03-028,Santa Fe Insti-

tute.

Hatakeyama, G., Kimura, K., Akiyoshi, M., and Komoda,

N. (2007). A programming environment for multi-

agent simulation based on graph representation. In

Proc. of 2007 Summer Computer Simulation Confer-

ence (SCSC 2007), in CD-Rom.

kke. Mas community. http://mas.kke.co.jp/index.php.

MacNealy, M. S. (1999). Strategies for Empirical Research

in Writing. Allyn Bacon.

Minar, N., Burkhart, R., Langton, C., and Askenazi, M.

(1996). The convergence and stability of cultures.

Technical report, Working Paper 96-06-042,Santa Fe

Institute.

North, M., Howe, T., Collier, N., and Vos, R. (2005). The

repast simphony runtime system. In Proc. of Agent

2005 Conference on Generative Social Processes.

A DESCRIPTION METHOD FOR MULTI-AGENT SIMULATION MODEL UTILIZING TYPICAL ACTION

PATTERNS OF AGENTS

319