EXPERIENCING AUML IN THE GAIA METHODOLOGY

Luca Cernuzzi

DEI Universidad Católica “Nuestra Señora de la Asunción”, Campus Universitario, C.C. 1683, Asunción, Paraguay

Franco Zambonelli

Dipartimento di Scienze e Metodi dell’Ingegneria, Università di Modena e Reggio Emilia, Reggio Emilia, Italy

Keywords: Agent Oriented Methodologies, Gaia, AUML, O

pen Multi-Agent Systems

Abstract: In the last few years a great number of AOSE methodologies have been proposed, some of which centered

on organizational aspects to better capture the behavior of agents societies. Those methodologies may be

considered very useful for modeling open systems composed of a great number of interacting autonomous

agents. Gaia exploits organizational abstractions to provide clear guidelines for the analysis and design of

complex and open Multi-Agent Systems (MAS). However, the notation of the Gaia methodology is

probably less powerful (and perhaps less acceptable for industry solutions) than others (like AUML). In this

perspective, this aims at performing a preliminary exploration towards the potential application of the

AUML notation into the Gaia methodology: it explores the above issues using an application example and

pays specific attention to the problem of modeling the complexity of open MAS and emergent behaviors.

1 INTRODUCTION

In the last few years a great number of AOSE

methodologies have been proposed trying to model

specific agents architectures, or extending accepted

techniques and methods from traditional OO

engineering paradigm. However, they are dependent

on abstractions and tools that may be unsuitable for

modeling new trends in agent-based systems..

Other researchers have recently proposed to

entify appropriate abstractions for Multi-Agent

Systems (MAS), and to propose software

engineering methodologies accordingly. Some of

such methodologies are Gaia (Zambonelli et al.,

2003), MESSAGE (Caire et al., 2001), TROPOS

(Giunchiglia et al., 2002), and ROADMAP (Juan et

al., 2002). All of these methodologies share the idea

that a MAS may be viewed as an organized society

of individual agents with their roles and different

kinds of interactions among them according to

specific protocols that are related to the roles of the

interacting agents. Some of those methodologies

introduce different abstractions however, very few

of them explicitly focus on organizational ones.

Among those, we consider that the new Gaia

roposal (Zambonelli et al., 2003) (the first proposal

was centered on closed communities of cooperating

agents) may be specially significant when used in

the analysis and design of open MAS. In effect, Gaia

exploits organizational abstractions that are

necessary for designing and building systems in

complex, open environments. Our considerations are

reinforced by the evaluation of Gaia as presented in

(Sturm and Shehory, 2003).

However, probably due to its simplicity, Gaia

notations are

poor and far to be widely accepted for

industry solutions (unlike UML in OO software

engineering). This aspect seems to be quite evident

in the specification of agent interactions (Sturm and

Shehory, 2003). In fact, the Gaia protocol model

while rigorously specify actors and input and output

of the protocol, use informal natural language to

specify its semantics, including the dependency and

speech-act interactions involved. Given this, it may

be very interesting to re-use richer notations founded

on a more consolidated paradigms, like object

orientation with its proven techniques and solutions,

and adapt them to MAS specification needs.

In this paper we try to apply AUML to Gaia.

UML is not per se a thorough methodology

However it models in a very rich and expressive way

the Agents Interaction Protocols (AIP) that

constitute a central aspect for open MAS. Thus, AIP

may naturally replace the Gaia protocols model.

283

Cernuzzi L. and Zambonelli F. (2004).

EXPERIENCING AUML IN THE GAIA METHODOLOGY.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 283-288

DOI: 10.5220/0002618802830288

 SciTePress

The remainder of this paper is organized as

follow. Section 2 introduces the idea of open MAS

and an illustrating example. Section 3 synthetically

presents the new Gaia proposal. Section 4 analyzes

the AUML notation and describes how it is possible

to use it within Gaia by means of the analysis

specification of the illustrating example. Section 5

discusses related work in the area. Finally, Section 6

concludes and outlines open research directions.

2 OPEN MAS: AN EMERGENT

PARADIGM

Agents may be conceived as stand-alone entities that

accomplish particular tasks on behalf of a user (e.g.,

personal digital assistants, e-mail filters, buy

assistants). However, usually, the environments

where agents accomplish their tasks are populated

with other agents. In these MAS, the global behavior

of the system derives from the interaction (co-

operation, co-ordination, negotiation, etc.) among

the existing agents.

So, a MAS is based on two types of aspects:

individual and collective. The former set comprises

those aspects found in classical systems but is the

latter that includes those constituting a new

dimension: the organizational aspects.

It may be possible to distinguish between two

main classes of MAS:

- distributed problem solving systems in which

agents are explicitly designed to co-operatively

achieve a specific goal in a closed way. That is, all

agents are defined a priori, they are co-operative to

each other and, therefore, they can trust one another

during interactions;

- open systems in which each agent has its own aims

and objectives and is not designed to share a

common goal with others. Moreover, since the

objectives of different agents may be in opposition

(competitive), agents should not necessarily trust

each other.

Actually, a growing trend in agent applications

focuses on open MAS based on a great number of

agents whose interactions may produce an emergent

behavior and distributed intelligence. Examples of

them may be found in most Internet-based systems

(e.g., agents for information retrieval and web

service agents) in which the agents have to exploit

services, knowledge, and capabilities offered by

other agents. Also, it includes all those systems that

involve interactions between agents on behalf of

different stakeholders (e.g., e-commerce agents).

Critical aspects in open MAS applications are the

organizational structure (if any) and specially the

organizational rules that control the behavior of self-

interested agents. For this reason it is very important

to focus on methodologies that may support the

modeling of those aspects, among the agents with

their roles and interaction.

2.1 Illustrating Example: the Agents

Marketplace

A representative example of open MAS are agent

marketplaces. A marketplace can be considered any

place where some proactive entities (e.g. persons,

enterprises, or computational organizations) can go

to put on sale services and/or goods and, vice versa,

where other proactive entities go to buy good and

service they are in need of.

Currently, marketplaces may exist in the form of

Web sites, where specific communities of users

interested in specific classes of goods can meet and

arrange their commercial transactions in an

interactive way and without the constraint of

physical co-location. Still, the need of some form of

interaction limits the capability and widespread

acceptance of such Web sites, due to the amount of

time which may be required for looking and

contracting for goods.

Agents can effectively accomplish the tasks of

looking for goods on behalf of clients, selling goods

on behalf of providers, and negotiating with each

other directly, without direct intervention of

involved humans and/or enterprises (apart from the

interaction with its agents). Thus, we can envision

the Internet will be populated with a variety of

special-purpose marketplaces.

In such marketplaces, agents interested in

specific classes of goods will meet to access an

environment made up of "sales offers" (possibly

based on an auction model) and of "wanted

requests". Such agents, in a given marketplace, will

form an organization made up of agents playing the

roles of "client" and "provider" in a wanted request

model as well as those of “bidder” and “supplier” in

an auction-based model, and interacting with each

other according to specific negotiation patterns. The

intrinsic openness of the scenario, where different

agents may enter the marketplaces to negotiate,

introduces the issues of controlling negotiations in a

proper way so as to avoid agents, which have a self-

interested behavior, cheat to with each other.

3 GAIA IN A NUTSHELL

A first overview of the Gaia methodology with its

models and their relationships is presented below.

ICEIS 2004 - INFORMATION SYSTEMS ANALYSIS AND SPECIFICATION

284

Figure 1: The Gaia Methodology

Gaia does not consider the collection of

requirements stage, starting right at the analysis

phase which aims to collect all the specifications of

the computational organization of the MAS. This

includes the identification of (see Figure 1):

- The goals of the organizations that constitute the

overall system and their expected global behavior.

This will be refined during the design phase in order

to identify useful decomposition of the global

organization into sub-organizations.

- The preliminary roles model that captures the

basic skills required for different types of agents to

reach the organization goals. At this stage, Gaia’s

notion of roles is abstract from any mapping into

agents (this issue will be considered in the design

phase) and the analyzer have to avoid the imposition

of a specific organizational structure.

- The preliminary interaction model that captures

the basic needed interactions from the identified

preliminary roles. Also this model, possibly not

being completely defined, must abstract away from

the organizational structure.

- The organizational rules that govern the

organization in its global behavior. Such rules

impose constraints on the execution activities of

roles and protocols. Moreover, they are fundamental

to efficiently specify the openness and the general

behavior of the developing MAS.

The output of the analysis phase consists of three

basic models: (i) a preliminary roles model; (ii) a

preliminary interactions model; and (iii) a set of

organizational rules.

The design phase is aimed at producing a

complete specification of the MAS. To this end, the

design phase contemplates the following sub-phases:

Requirements statements

Organizational

Rules

Preliminary

Interaction Model

Preliminary

Role Model

Interaction

Model

Service Model

Role Model

Agent Model

ANALYSIS

FHASE

DESIGN FHASE

Catalogue of

Organizational

Patterns

Organizational

Structure

- Definition of the overall organizational structure

considering the organizational rules. At this stage it

is important to exploit well-known organizational

patterns that may also influence the design of the

final interactions model.

- Considering the adopted organizational structure

a revision and completion of the preliminary role

and interaction models.

- Definition of the agent model specifying agent

types (a set of agent roles) and agent instances.

- Definition of the services model which specify

the main services (blocks of activities with their

conditions) related with the agent roles.

The specification of agents with their roles and

the interactions among them and with the

environment, are not enough to capture the complex

and emergent behavior derived from many self-

interested agents applications. For this reason, Gaia

spent additional effort in modeling the

organizational structure as well as the organizational

rules. Those aspects, as already stated in section 2,

are very relevant in order to specify open MAS.

4 INTEGRATING AUML INTO

GAIA

Different attempts in the past few year have tried to

extend UML notation for agent-based systems one

of them is Agent UML (AUML). AUML is not per

se a thorough methodology, however it builds on the

acknowledged success of UML in supporting

industrial-strength software engineering.

The core part of Agent UML is the Agents

Interaction Protocol (AIP), that constitute a central

aspect for open MAS, specified by means of

protocol diagrams. Protocol diagrams extensions to

UML include agent roles, multithreaded lifelines,

extended message semantics, parameterized nested

protocols, and protocol templates.

Still, the key ideas of AUML that may be

integrated into Gaia methodology are:

- The protocol can be regarded as a whole entity

(expressed by mean of an enriched sequence

diagram) and treated as a package. AUML considers

an AIP as a template, whose parameters may be

roles, constraints, and communication acts. This

template approach expresses in a more compact way

and UML-like notation the same semantics of the

Gaia protocol notation, but it is easier to visualize.

- Each protocol implies inter-agent interactions

that are described using sequence diagrams, activity

diagrams, and statecharts. AUML extends sequence

EXPERIENCING AUML IN THE GAIA METHODOLOGY

285

diagram notation in order to represent Agents (and

eventually their Class) and their Roles, and to

support concurrent threads of interactions. The

activity diagram, particularly useful for complex

interaction protocols that involve concurrent

processing, and statecharts are used to specify the

internal behavior of an agent.

AUML proposes other extensions to UML such

as packages with agent interfaces, deployment

diagrams indicating mobility, emergence, etc.

However, those notations are less richer than those

proposed for AIP and some of them are poor

compared to Gaia notations. For example, the role

specification in Gaia is more expressive, formal and

includes more relevant aspects (permissions and

responsibilities) than AUML proposal.

Moreover, AUML does not cover all the

abstractions proposed by Gaia. Specifically, AUML

(Parunak and Odell 2002) offers a rather poor

notation in covering the organizational structures

and does not consider the organizational rules. Thus,

it presents some barriers to adapt to complex and

open systems with self-interested behavior.

4.1 The Agents marketplace modeled

in Gaia + AUML

In this section the illustrating example of Agents

Marketplace is specified using Gaia and replacing

the preliminary interaction model with the AUML

notation. For space reasons we present just a few

examples for each model of the analysis phase in

which it is possible to appreciate the main benefits

of the integration.

The Organization

In the example of the agents marketplace, the need

to request goods or services usually implies the

request for a set of offers by sellers, and receipt of

the offers, and the evaluation by the buyer, after

which the service provision is assigned to the

winner. Such a solution can be easily delegated to a

MAS with the same organizational structure.

However, it could also be possible to adopt a

different structure. For instance, to improve

efficiency one can adopt a descending price auction

for requesting services at the best price. The choice

for an auction-based negotiation requires re-thinking

the organizational structure and, for instance,

introduces the need for agents to interact with the

mediation of an auctioneer in charge of enacting

negotiation rules. Moreover, it is possible to

conceive several interacting organizations to co-exist

in a marketplace. For example, one could think of

two separated organizations: one for dealing with the

contracting phase and another for dealing with the

subsequent payment and delivery phases. In the

next, we pay attention just to the organization for

dealing with the contracting phase.

The Preliminary Role Model

In the agents marketplace example, it is possible to

identify five possible roles corresponding to three

classes of agents: Client and Bidder (for the agents

class Buyers); Provider and Supplier (for the agents

class Sellers); and Auctioneer (for the agents class

Auctioneers, see Figure 2).

Role Schema: Auctioneer

Description: Mediator between supplier and bidder in the

“Auction ” model (the role of the agents class Auctioneers).

Protocol and Activities:

ServiceProposed, Offers,

ReceivePriceOffers,

PriceEvaluation,

AcceptPrice, AskForNewBid, Inform

Permissions:

Reads offer_definition //the offer made by the seller

Changes price //the highest proposed

price

Responsibilities:

Liveness:

Auctioneer=

(ServiceProposed.Offers.ReceivePriceOffer

PriceEvaluation.AcceptPrice!AskForNewBi

Safety

 number_of_price_proposal >= 1

Figure 2: Schema for role Auctioneer

Any role schema is intended to be a semiformal

description of an agent’s behavior and specifies

permissions and responsibilities corresponding to

that role. Responsibilities may be specified in terms

of liveness (desirable) and safety (avoided

undesirable) properties expressed using regular

expressions. Those expressions include a set of

activities (actions that the agent may perform

without any interaction with other agents) and

protocols (activities that do require interaction

among agents). For example, in Figure 2 an

Auctioneer needs to access the offer presented

by a seller and to propose the seller the highest price

offered by bidders, as stated in its permissions. The

liveness expression, that may occur 0 or more times,

specifies that whenever the Auctioneer

receives a

proposition of a service (by means of the

ServiceProposed protocol), then Offers this

proposition to Bidders,

ReceivePriceOffers from Bidders, and

then may AcceptPrice or AskForNewBid. The

safety expression states that the Auctioneer

needs at least one price proposal.

ICEIS 2004 - INFORMATION SYSTEMS ANALYSIS AND SPECIFICATION

286

The preliminary interaction model

This model consists of a set of protocol definitions,

one for each protocol of each role in the system.

More attention is paid to the nature and purpose of

the interaction than to the sequence of execution

steps. In fact, the protocol definition describes the

high-level purpose of the protocol, ignoring

implementation details such as the sequence of

messages exchanged. In particular, the protocol

definition is a simple table detailing the role

initiating the protocol, the role in charge of

responding to it, the input and output information

processed in the protocol, as well as a brief textual

description of the type of information processing

taking place during the execution of this protocol

The proposed notation of Gaia for protocols is

quite informal. Thus, instead of the Gaia notation,

we introduce the use of AIP notation proposed by

AUML. This implies that a protocol must be

described as a sequence of actions and message

interactions and may contain a set of atomic

protocols as defined in Gaia. In the example we have

considered two protocols for the contract phase: one

for the "Wanted Request" model and one for the

"Auction" model. For space reasons we present just

the packagefor "Auction" protocol (see Figures 3),

avoiding the activity diagram and statechart.

It states that when the Supplier proposes a

service, the Auctioneer offers it to the Bidder

looking for the best price. Meanwhile, a Bidder may

present a price offer or may inform the Auctioneer

that he has not understood the proposal. Once the

Auctioneer has evaluated the price it may accept it,

informing the Supplier, or reject it asking for new

bid.

The Organizational Rules

Gaia considers the Organizational Rules in a

perspective that is coherent with responsibilities

characteristics of roles but referred to the

organization as a whole. Accordingly, it is possible

to distinguish between safety and liveness

organizational rules. As an example of a safety rule,

the Auctioneer cannot participate as a Bidder in an

auction it is moderating:

⌐Auctioneer (offers (x)) ⎜ Bidder (offers (x))

Also, in the case of a negotiation based on the

English auction mechanisms, bidder agents must not

be allowed to interact directly with the seller, so as

to avoid collusions aimed at artificially heightening

the selling price of a good.

4.2 Discussion

A possible weakness of Gaia is the notations it

proposes and, among them as pointed out in (Sturm

and Shehory, 2003), specifically that related to the

protocol model. In fact, the proposed notation

considers all the relevant aspects of a protocol but

may be too extensive to specify (one model for

every interaction), and it is quite informal and not

based on a standard accepted by industry.

As previous section highlights, the integration of

AUML within Gaia leads to a richer notation for the

specification of protocols and inter-agent

interactions since the AUML notation introduces

different advantages. First, it specifies a

distinguished set of agent instances satisfying the

agent role and class it belongs to; while Gaia just

Figure 3: The Auction model protocol

EXPERIENCING AUML IN THE GAIA METHODOLOGY

287

specifies the role. Second, it is more compact

specifying in a single diagram a sequence of actions

and messages interactions which may contain a set

of atomic protocols as defined in Gaia. Third,

AUML is more formal and let to specify the time

ordering of messages between agents. Finally,

AUML notation introduces the opportunity for

agents to select a path in the interaction according to

their goals. The latest two aspects are described in

Gaia using natural language and so introducing

possible ambiguities and misunderstandings.

5 RELATED WORK

Other attempts to extend Gaia for better modeling

open MAS include Roadmap (Juan et al., 2002) and

Skeleton (Juan et al., 2003) methodologies.

The Roadmap methodology aims to support the

engineering of large-scale open systems promoting

the view of software systems as computational

organizations. Roadmap extends the Gaia

methodology by introducing use-cases for

requirement gathering, explicit models of agent

environment and knowledge, and an interaction

model based on AUML interaction diagrams (Juan

et al., 2002). However, the interaction model

proposed is just a statement without further details

or examples and it seems to be an interesting

possible idea more than a real conceptualized model.

For this reason, it is quite unclear for designers how

to accomplish this integration.

The Skeleton methodology proposes an

integration of the common elements identified from

Prometheus (Padgham and Winikoff, 2002) and

Roadmap. It inherits from Roadmap the interaction

model with its advantages and the same drawbacks

mentioned above.

6 CONCLUSIONS AND FUTURE

WORKS

This work proposes the integration of AUML within

the Gaia methodology to improve modeling of open

MAS. Specifically, we replaced the protocol model

of Gaia with the Agents Interaction Protocol (AIP)

of AUML, specified by means of protocol diagrams.

This extensions to UML enrich Gaia in four main

aspects: (i) a richer notation for specifying agent

instances of a particular class satisfying the agent

role; (ii) a more compact notation that represents in a

single diagram a sequence of actions and message

interactions; (iii) a more formal notation that reduces

possible ambiguities and allows to specify messages

between agents; and (iv) multithreaded lifelines that

permit agents to select a path in the interaction

according to their goals.

Moreover, AUML builds on the acknowledged

success of UML in industrial software engineering

and it is reasonable to think that it may reduce the

distance between researchers’ proposals and industry

practices. Nevertheless, the main pitfall in using

Gaia integrated with AUML to design MAS in an

industrial environment is that there are no CASE

tools available which implement this methodology.

REFERENCES

Caire, C., Garrijo, F., Gómez, J., Pavón, J., Leal, F., et al.,

“Agent oriented analysis using MESSAGE/UML”.

Proceedings of Agent-Oriented Software Engineering

– AOSE 01, Montreal Canada, May, 2001

Giunchiglia, F., Mylopoulos, J., and Perini A., “The

Tropos Software Development Methodology:

Processes, Models and Diagrams”, Proceedings of

Agent-Oriented Software Engineering (AOSE-2002),

Bologna, Italy, July 2002

Juan, T., Pearce, A. and Sterling, L., “ROADMAP:

Extending the Gaia Methodology for Complex Open

Systems”. Proceeding of Autonomous Agents and

Multi-Agent Systems - AAMAS ’02 (pp. 3-10),

Bologna, Italy, July 15-19, 2002

Juan, T., Sterling, L. and Winikoff, M., “Assembling

Agent Oriented Software Engineering Methodologies

from Features”. Autonomous Agents and Multi-Agent

Systems-AAMAS‘03, Melbourne, Australia, July, 2003

Odell, J., Parunak, H. v. D., and Bauer, B., “Extending

UML for Agents”. Proceedings of Workshop on Agent

Oriented Information Systems – AOIS 2000, Austin,

USA, 2000

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

Methodology for Developing Intelligent Agents”.

(poster) Proceedings of Autonomous Agents and

Multi-Agent Systems - AAMAS ’02 (pp. 3-10),

Bologna, Italy, July 15-19, 2002

Parunak, H. v. D., and Odell, J., “Representing Social

Structures in UML”. In: Wooldridge M., et al. (eds.):

Agent Oriented Software Engineering – AOSE II (pp.

1-16), Springer-Verlag, Berlin, Germany, 2002

Sturm, A. and Shehory, O., “A Framework for Evaluating

Agent-Oriented Methodologies”. Agent Oriented

Information Systems – AOIS 2003 at AAMAS ‘03,

Melbourne, Australia, July 14, 2003

Zambonelli, F., Jennings, N., and Woolddridge, M.,

“Developing Multiagent Systems: the Gaia

Methodology”. ACM Transactions on Software

Engineering and Methodology, 12(3) (pp. 417-470),

July 2003.

ICEIS 2004 - INFORMATION SYSTEMS ANALYSIS AND SPECIFICATION

288