AOPOA

Organizational Approach for Agent Oriented Programming

Enrique González, Miguel Torres

Grupo de Investigación SIDRe, Pontificia Universidad Javeriana,Bogotá, Colombia

Keywords: Agent-Oriented Programming, Agent Design Methodology, MultiAgent Systems.

Abstract: This paper presents AOPOA, an agent oriented programming methodology based in an organizational

approach. The resulting multiagent system is composed by a set of active entities that aim to accomplish a

well-defined set of goals. This approach allows to design complex systems by decomposing them into

simpler ones. The organizational approach makes it easier to perform an iterative and recursive

decomposition based in the concept of goal; and at the same time to identify the interactions between the

entities composing the system; at each iteration an organization level is developed. During the analysis

phase, tasks and roles are detected. During the design phase, the interactions are characterized and managed

by means of cooperation protocols. At the final iteration, the role parameterization is performed, which

allows to specify the events and actions associated to each agent. Finally, the deploy of the agent instances

is determined allowing redundancy to achieve the requirements of the system.

1 INTRODUCTION

The rational agent concept appeared in AI as a new

conceptual and practical approach for designing and

building intelligent systems. Rational agents are seen

as units acting to attain a set of well-defined goals.

In a more general approach, an agent appears as

entity encapsulating data, knowledge and behaviour,

capable to perform a task in an autonomous and

proactive way. Normally, agents do not work alone,

but form groups to attain global goals, which can not

be achieved by one agent alone; this group of

cooperating agents form a MultiAgent System

(MAS). MAS have appeared as a new way to build

systems that solve complex problems.

Building a MAS implies a different vision of the

design process, it must include and take advantage

of the agent’s intrinsic characteristics. A potential

advantage in applying the Agent Oriented

Programming (AOP) paradigm is that it facilitates

the development of complex applications. This

design approach is well-suited to subdivide a

complex problem into simpler ones, which are

solved by active entities, the agents; besides, it

offers the possibility to design a modular solution

that allows a more structured and coherent

management of global system complexity. However,

it is clear that when dividing the system, there are

multiple problems to solve to get the agents to work

together in a cooperative way in order to fulfil the

system’s goals. Thus, AOP must provide methods

not only to assign responsibilities to agents, but also

to identify and manage the interactions between

them.

The development of methodologies to perform in

an efficient way the analysis and design of an agent

based system is a research field still open; even if

there are some already proposed alternatives. This

paper presents AOPOA, an AOP methodology based

on an organizational approach. In this approach, a

MAS is perceived as an organization. As in other

previous approaches, complexity management is

attained through an organizational decomposition of

the system in simpler parts. The key point of the

approach of AOPOA is that at the same time agents

are designed in a structured and progressive way,

relationships between them are automatically

established and characterized.

González E. and Torres M. (2006).

AOPOA - Organizational Approach for Agent Oriented Programming.

In Proceedings of the Eighth International Conference on Enterprise Information Systems - SAIC, pages 75-80

DOI: 10.5220/0002453500750080

 SciTePress

2 AOP METHODOLOGIES

The problem of building a system based on an AOP

methodologies has already been studied

(Wooldridge M. 2000) (Alonso F. 2004). In the

development process of AOPOA some existing

methodologies were taken into account, and they

were analyzed to find the more relevant

characteristics that a good methodology must posses.

There are a great variety of methodologies for

designing Multiagent Systems, some of them are

extensions or are based in other design models or

methodologies, inheriting its benefits and its

failures. Alonso et.al (Alonso F. 2004) proposed a

taxonomy to organize such methodologies in three

categories: methodologies based upon the Object

Oriented paradigm; methodologies based on Agents

itself; methodologies based on Knowledge

Engineering.

The methodologies based upon the Object

Oriented Software paradigm have certain problems

such as a generic analysis model, and most of them

do not cover the social structure of the system and

the environment characteristics. Under these

methodologies the agent is a complex object,

reducing the level of real abstraction provided for

agents. Such methodologies also use some models

and views (i.e. UML diagrams) and some of the

techniques proposed in Software Engineering

common processes.

The methodologies based on Agents are based

upon abstraction of social levels such as groups and

organizations. These methodologies are strong in the

first steps of the specification and design level of the

MAS, but as in the previous methodology they lack

of a generic analysis model which can be used to

assess if any given MAS approximation is

appropriate for the given problem. These

methodologies also present different levels of

abstraction of the MAS such as: the internal

structure of the agents, the structure of the

interactions among agents, and the social structure

of the different groups of agents.

Finally, the methodologies based on Knowledge

Engineering are characterized by the identification,

acquisition and modeling of the knowledge used by

the agents in the MAS. The most representative

methodologies of this category are extensions of the

CommonKADS (Schreiber G. 1999) methodology

for developing Knowledge Based Systems. For

instance, MASCommonKADS also appropriates

some object oriented design and analysis techniques.

In next paragraph some of the most relevant MAS

methodologies are presented.

Tropos is an agent oriented methodology

(Penserini, L. 2004), is based upon two basic

concepts: the notion of an agent who uses plans in

order to fulfill goals and the covering of the early

and late requirements analysis. Prometheus

(Padgham, L. 2002) is detailed and complete, and

covers all the steps since the requirement analysis

process until the MAS implementation using the

JACK framework (Howden N. 2001). Odell presents

a methodology based on the object oriented software

paradigm, and it uses as a basis the UML diagrams

for the MAS representation (Odell J. 2004); using

Metamodels in order to describe the MAS, and its

elements. GAIA (Wooldridge M. 2000) is another

object oriented based methodology, which uses the

initial concept of organization and sub organization,

modeling of the environment of the MAS, role

modeling and interaction model among roles. GAIA

does not present particular techniques for

implementation or for requirements elicitation.

MASE is a MAS development methodology

(DeLoach S. 2000), which goes from initial

specification until the actual implementation of the

MAS. The process of capturing goals produces a

goal hierarchy that is used to also identify Use

Cases; which are then used to generate sequence

diagrams among roles.

To summarize, the fundamental characteristics in

the development of a methodology are: to identify

the goals that must be attained in order to solve the

problem; to assigning them to roles, which will

perform the necessary tasks to achieve the

objectives; and then, to establish the social aspects,

such as interaction and cooperation mechanisms

required to get the desired social behaviour; finally,

the assignation of the roles to the agents which will

conform the system. In order to achieve this process,

there are different alternatives, as has been shown.

3 MULTIAGENT SYSTEMS

An agent can be defined as an entity that perceives

its surrounding environment through sensors, and

which also responds or acts in that environment

through effectors (Rusell N. 2003). Agents respond

to events coming from the environment or from

other agents; agents select the most adequate action

that leads the agent to achieve its own goals.

ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING

A MultiAgent System (MAS) is a set of organized

agents; they interact in a cooperative way to reach in

a collective way the global goals of the system

(Ferber J. 1999). A MAS can be viewed as a

organization of agents in which interactions are the

origin and product of the system’s persistence and

dynamics through time. Cooperation is an important

issue in agent interactions, and it is composed by

three characteristic elements: collaboration,

coordination and conflict resolution. Collaboration is

required when the agent’s abilities and resources are

not enough for the agent to accomplish goals.

Coordination is related to the order in which the

system’s tasks must be performed. When resource

conflicts arise, they must be solved; usually agents

have to negotiate or to apply rules that will impose

certain social restrictions. Finally, agent

communication is the support for cooperation. In

practice, any cooperation technique can be modelled

by an interaction protocol, which defines an ordered

set of communication acts between the implicated

agents.

In a MAS, a role makes reference to an abstract

entity, whose function is to achieve a set of goals; in

other words, a role defines a set of tasks. The

accomplishment of such a task depends on the

abilities, resources and bindings between the entity,

its environment and other roles. The objective then

is to build an efficient system, in which the abilities

and resources of a role are as different as possible

from others, in order to avoid redundancy of abilities

and reduce resource conflicts.

An organization is an array of relationships among

individuals; that can be perceived as a single unit or

system. The organizational structure is defined in an

abstract way by a set of roles, which can be assumed

by instantiated agents, and a set of relationships

between such roles. Each organization can be

perceived as a set of organizations; each of them can

be decomposed in a recursive way into lower-level

organizations.

4 GOAL DECOMPOSITION

In the AOPOA methodology, there are two main

processes: analysis and design; each one of them is

applied iteratively until it is determined that it is no

longer necessary to decompose the goals and roles

already identified. Based on these organizational

perspective, the system is viewed as an organization

that can be recursively divided. In each iteration, the

analysis of an organizational level is performed. In

this way, a MAS can be modelled as an

organizational tree, where the root node represents

the whole system, child nodes represent progressive

role decompositions, and leaf nodes represent final

roles that can be instantiated as the agents of the

functional system. The iterative process of recursive

decomposition ends when it is considered that the

complexity of all the final roles is low enough. Once

this point is reached, a final iteration is performed, to

make the detailed design of the agents that will

represent the existing roles in the final system.

The key concept for identifying a role is the

association of an autonomous entity to a set of goals.

In AOPOA, the notion of goal is the foundation,

which allows the decomposition process of a

complex role into simpler ones. In fact, this sub-role

generation process implies two activities:

- decomposition of the associated goals of a role

into simpler ones.

- grouping this new sub-goals to create new less

complex roles.

The process begins by modelling the MAS as a

single first level unique role. The first role’s goals

and environment interactions are derived directly

from the requirements that where identified for the

entire system. This general system’s goals must be

divided into sub-goals, which can be assigned to less

complex derived roles.

Another AOPOA key concept is the use of the

notion of interaction. Two agents interact when they

join efforts and abilities to reach a goal; or when

they have to synchronize its activities; or when

agents share resources during the execution of a

task. Every interaction is represented by an abstract

relationship, between the involved roles, called

cooperation bind. During the iterative decomposition

phase, the identification of interactions by goals is

performed directly. Every time a complex goal is

divided, if some of the resulting roles include a sub-

goal of the former one, these roles will probably

need to interact in order reach the complex goal.

Also through the decomposition, the identification of

the resources required by the new derived roles is

performed; when a shared resource is detected, the

implicated roles will probably need to interact.

As can be observed, different cooperation binds

automatically arise because of the potential

interactions that can appear between roles.

AOPOA - Organizational Approach for Agent Oriented Programming

During the design phase, every cooperation bind is

analysed; depending in the type of interaction, an

adequate cooperation technique can be chosen. As

was previously explained, the implementation of a

cooperation technique is made through an

interaction protocol. In summary, from the bind

identification, it is possible to determine, in a

structured way, the adequate sequence of

interchanged messages between agents.

An important characteristic of cooperation binds is

that through the role decomposition, the associated

binds of a role are inherited by the resulting sub-

roles. In fact, the binds existing between roles in an

organizational level should be assumed by the sub-

roles in the next organizational level. Figure 1

illustrates the previous concepts; The roles are

represented as circles, and binds among them are

arrows. In iteration i there are two identified roles. In

iteration i+1 the sub-roles are included inside the

roles identified in the previous level. In iteration i

only two binds where identified, and in iteration i+1

those binds remain and are extended to some of the

new sub-roles. New interactions among roles in the

same organizational level can also appear, for

instance the bind between roles 11 and 12.

Figure 1: Inheritance of cooperation binds.

5 AOPOA PHASES

Requirements are the basis upon which the system’s

general goals are identified. Once the problem is

well understood, it is examined in order to decide if

an agent-oriented approach is well-suited for the

problem at hand.

5.1 Analyse Phase

In the first iteration, the general objectives of the

system are analysed and divided. In each iteration,

tasks are defined for each goal to reach; a task is

characterized by the set of resources and abilities

required to attain the goal. Based on these tasks, a

grouping process is performed to determine the ideal

set of required roles; each group of tasks is

assimilated to a new role. Notice that only the tasks

derived from a role are grouped to obtain its

corresponding sub-roles; as a consequence, the

grouping procedure is applied to small sets of tasks.

The grouping process aims to qualify and group

tasks applying three different criteria:

- No opposite roles are assigned to the same role.

- There are few resource conflicts among the new

generated roles.

- The presence of abilities’ redundancy should be

reduced among roles.

The procedure to find the evaluate a generated

grouping is as follows:

1. Assign tasks in a way that a set of groups NG can

be produced. Each group of tasks represents a

possible candidate role to create. NG is the total

number of groups identified.

2. Evaluate the criteria for each group of tasks. The

applied criteria for a group g are: Aog, Ahg and Arg;

which respectively measure the criteria for goals,

abilities and resources. In order to perform this

evaluation equation 1 is applied.

(Eq. 1)

Ajg is the classification value of the indicator for a

criteria j for the evaluated group of tasks g; j can

have the values: h, o and r; g is the identifier of the

group that is being evaluated, and its value ranges

from 1 to NG.

NIjg is the number of intersections of type j; if j=o

the amount of goals of the different type in group g,

ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING

if j=h the quantity of redundant abilities that are

present in other groups; if j=r the required shared

resources that are present in other groups of tasks.

Njg is the total number of elements of type j inside

group g, i.e. objectives, abilities and resources inside

the group of tasks included in g.

3. Calculate an average Pj for each criteria j by

using equation 2.

∑

jgj

(Eq. 2)

4. Calculate a weighted sum of Po, Ph, and Pr.

An optimisation technique is used to find the

grouping with minimum evaluation value. This

process is not to long, as only few tasks are

considered at the same time. Finally, in order to stop

the decomposition procedure, an evaluation of the

complexity associated to each new role is

performed.

5.2 Design Phase

For each iteration, once the roles have been

identified during the analysis phase, the design

phase starts, and cooperative binds among roles are

identified. Once binds are identified, they are

characterized as interaction situations were the

coordination, collaboration and conflict resolution

techniques are present. For each identified

interaction the most suitable method must be

applied. It is open for the developer to use already

available methods and protocols or its own ones.

Finally, the selected protocols are translated into

communication binds.

5.3 Final Iteration

Once the analysis of complexity of the roles shows

that there is no need for any new division of roles,

then the final iteration of the AOPOA methodology

is applied. In this last iteration, the set of resulting

leave roles in the last level of the branches in the

organizational tree is parameterized, thus obtaining

the agent prototypes. A prototype of an agent has a

specification of goals, sensorial inputs, effector

outputs on the environment, and a definition of the

actions that the agent can perform. Once the agents

are specified, the events that trigger the actions

performed by the agent must be analysed, also the

mental abilities to execute these actions are

specified. Events can be perceptions from the

external elements in the environment or messages

received associated to the interaction communication

protocols. For each incoming event, the actions that

must be taken must be determined.

In Figure 2 the complete process is depicted, the

stages of the AOPOA methodology are shown.

Circles in the diagram represent artefacts generated

during this process.

Figure 2: Phases of the AOPOA methodology.

6 DISCUSSION

The identified advantages of using AOPOA are:

- The organizational approach helps to deal with

the system complexity and also to divide the

problem in smaller ones with well-defined

relationships.

- The grouping procedure aims to find a good

combination of task groups, in order to optimize the

role generation process.

AOPOA - Organizational Approach for Agent Oriented Programming

· The process is complete, since considers all the

required aspects to develop a good MAS; starting

with requirements elicitation until defining the

communication protocols and agent instances.

The selected case of study to test the AOPOA

methodology, was a based on the construction of a

restaurant simulation. The case of study also

provided a way to test the methodology’s

organizational, intra-agent and intra-organizational

scalability. Organizational scalability implies that a

system can be designed to be part of a greater

system, i.e. a restaurant can be part of a food chain.

Intra-agent scalability, means that new objectives

can be added to an already existent agent role.

Finally, intra-organizational scalability allows to

aggregate new roles to different organization’s

levels, over an already existent system. A detailed

explanation of the case of study is out of the scope

of this paper.

In order to implement the case of study, the BESA

agent framework was used (González E. 2003). The

AOPOA model transformation into a BESA

implementation was direct and fast. The AOPOA

model allows a rapid and robust event and action

implementation of the MAS. A detailed presentation

of the restaurant simulator design using AOPOA and

implementation using BESA can be found in the

work of Ahogado and Reinemer (Ahogado D. 2003).

Actual work to extend the AOPOA methodology

include:

The use of dynamic roles, as a mechanism for

agents to perform different roles accordingly to its

own objectives and situation.

Agents mobility, applied for dynamic agents who

can migrate through different machines in a

distributed system.

Taking into account the obtained results, it can be

concluded that AOPOA is a good choice for

constructing complex agent based systems. In fact,

the obtained advantages are derived from the

cooperative rational agent concepts used, allowing a

higher semantic level of the system and its

conforming entities.

REFERENCES

Ferber J., 1999. "Multiagent Systems: An Introduction to

Distributed Artificial Intelligence", Addison – Wesley

Longman. 1ra. Ed., 1999.

Rusell N., 2003. “Artificial Itelligence: A Modern

Approach”, 2

Edition, Pearson Education.

Ahogado D., Reinemer A.M., 2003. “Programación

Orientada a Agentes: Metodologías de Desarrollo de

Software”, Proyecto de Grado Carrera de Ingeniería de

Sistemas, Universidad Javeriana Bogotá.

González E., Bustacara C., Avila J., 2003. “BESA:

Behavior- oriented, Event-driven Social-based Agent

framework.”. PDPTA’03, Las Vegas-USA, CSREA

Press, vol. 3, Junio 2003, pp 1033-1039.

Alonso, F., et.al., 2004. Sonia: A methodology for natural

agent development. ESAW’2004 - 5th Intl.Workshop

on Engineering in the AgentsWorld

Wooldridge M., Jennings N., Kinny d., The GAIA

Methodology for Agent-Oriented Analysis and

Design, Autonomous Agents and Multi-Agent

Systems, vol. 3, pp. 285-312, Kluwer Academic

Publishers, 2000.

DeLoach, S. A. and Wood, M. F. 2000. Multiagent

systems engineering: the analysis phase. Technical

Report AFIT/EN-TR-00-02, Air Force Institute of

Technology.

Odell, J., Nodine, M. H., and Levy, R. 2004. A metamodel

for agents, roles, and groups. In AOSE, pages 78–92.

Howden, N., Rönnquist, R., Hodgson, A. and Lucas, A.

2001. JACK Intelligent Agents – Summary of an

Agent Infrastructure. The agent oriented software

group (www.agt-software.com.au).

Padgham, L. and Winikoff, M. 2002. Prometheus: A

methodology for developing intelligent agents. In

Proceedings of the Third International Workshop on

AgentOriented Software Engineering, Bologna - Italy.

AMMAS.

Penserini, L.; Kolp, M.; Spalazzi, L.; Panti, M. 2004.

Socially-based design meets agent capabilities.

Proceedings of the IEEE/WIC/ACM International

Conference on Intelligent Agent Technology

(IAT’04).

Schreiber, G., Akkermans, H., Anjewierden, A., de Hoog,

R., Shadbolt, N., Velde, W. V. D., and Wielinga, B.

1999. Knowlede Engineering and Management: The

CommonKADS Methodology. MIT Press.

ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING