HOW TO BUILD A MULTI-MULTI-AGENT SYSTEM
The Agent.Enterprise Approach
Tim Stockheim
Institute of Information Systems, Universität Frankfurt,
Mertonstr. 17, D-60325 Frankfurt am Main, Germany
Jens Nimis
Institute for Program Structures and Data Organization, Universität Karlsruhe (TH),
Am Fasanengarten 5, D-76131 Karlsruhe, Germany
Thorsten Scholz
Center for Computing Technologies (TZI), Universität Bremen,
Postfach 33 04 40, D-28334 Bremen, Germany
Marcel Stehli
Institute of Information Systems, Technische Universität Ilmenau,
Postfach 100 565, D-98684 Ilmenau, Germany
Keywords: Agent-oriented software engineering, Multi-agent systems, Application integration
Abstract: The maturity of technical foundations for multi-agent systems and the support by development tools, infra-
structure services, and a number of development methodologies leads to an increasing number of existing
multi-agent systems. A more and more networked environment drives the demand for coupling these het-
erogeneous systems to large multi-multi-agent systems. Unfortunately, the design and implementation steps
necessary in this context are currently not supported by established development methodologies; conven-
tional approaches mainly focus on isolated multi-agent systems. In this paper, we present an approach for
the integration of heterogeneous multi-agent systems. The Agent.Enterprise system is a coupled multi-
multi-agent system that has been designed and tested in the manufacturing logistics domain.
1 INTRODUCTION
With the growing success of FIPA-standardization
(http://www.fipa.org) for multi-agent systems
(MAS) and the increasing availability of FIPA-
compliant frameworks (e.g. JADE -
http://jade.cselt.it) various MAS have been devel-
oped and real applications of MAS are emerging.
These applications usually focus on a specific issue
within a certain domain (e.g. manufacturing). In re-
ality these systems often depend on each others’ in-
and output; therefore, the need for integration re-
spectively coupling of these systems arises. Thus,
coupled systems result in a large heterogeneous sys-
tem. In this context we refer to the concept of cou-
pled MAS as multi-multi-agent systems (MMAS).
Each MAS represents a closed organizational entity
with predetermined boundaries of its agents’ influ-
ence. The MMAS restricts the communication be-
tween single MAS to prevent uncontrollable and
unmanageable system complexity.
Nowadays, open service infrastructures such as
Age
ntcities (http://www.agentcities.org) as well as
technical foundations for the coupling of MAS are
available. However, guidance in the software engi-
neering process (analysis, design, implementation,
testing) for these MMAS is missing, since existing
364
Stockheim T., Nimis J., Scholz T. and Stehli M. (2004).
HOW TO BUILD A MULTI-MULTI-AGENT SYSTEM - The Agent.Enterprise Approach.
In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 364-371
DOI: 10.5220/0002638703640371
Copyright
c
SciTePress
agent-oriented development methodologies are fo-
cusing on isolated MAS only.
This paper presents an approach for analysis, de-
sign and implementation of MMAS that has been
developed and applied by five research projects
(Special interest group) for coupling their MAS. The
resulting system is called Agent.Enterprise and was
introduced in (Frey et al., 2003).
Outline. Section 2 considers and compares sev-
eral existing MAS development methodologies. On
this basis, we introduce the Agent.Enterprise ap-
proach as well as technical improvements (Gateway-
Agent-Concept), which set up basic conditions for
the approach in Section 3. The pros and cons of the
proposed approach and conventional methodologies
are discussed in Section 4. Finally, we conclude the
paper in Section 5.
2 AGENT-ORIENTED
DEVELOPMENT METHODS
Over the last years, the need for applicable and
broadly accepted development methods for multi-
agent systems resulted in a large number of efforts in
order to overcome this problem. Various methods
exist, which support at least one of the development
phases (analysis, design, implementation, and de-
ployment) with representations of varying formal
accuracy and of varying semantic foundations, e.g.
Gaia (Wooldridge, Jennings and Kinny, 2000),
PASSI (Costenio, and Potts, 2002), MASSIVE (Lind
2001), MaSE (Wood, and DeLoach, 2001), AUML
(Odell, Parunak and Bauer, 2001).
The focus of most of the mentioned methods is on
building single (most often closed) MAS and thus,
none of them considers the development or integra-
tion of MMAS. Nevertheless, one can be expect, that
the development process for MAS and MMAS will
have some joint properties. The following subsec-
tions take a closer look at three of the most promi-
nent methods in order to get a better understanding
of the nature of MAS development methods before
we present our method to interconnect different
MAS.
2.1 Gaia
One of the best known representatives is the Gaia
methodology for agent-oriented analysis and design
(Wooldridge, Jennings and Kinny, 2000). It starts
with a statement of requirements and ends up in a
sufficiently detailed system that can be implemented
directly. Hence, this methodology that consists of
analysis and design steps can be thought of as a
process of developing increasingly detailed models
of the system to be constructed.
Analysis. The objective of the analysis stage is to
gain an understanding of the system and its struc-
ture. The system's organization is viewed as a collec-
tion of roles that have certain relationships to each
other, and that take part in systematic, institutional-
ized patterns of interaction with other roles. Thus,
the organization model in Gaia comprises two fur-
ther models: the role model and the interaction
model.
The role model identifies the key roles in the sys-
tem. Such roles are characterized by two types of at-
tributes:
The responsibilities of a role: A role captures a
specific functionality. This functionality is repre-
sented by attributes known as the role's liveness
and safety responsibilities.
The permissions/rights associated with a role: A
role has certain permissions, relating to the type
and the amount of resources that can be exploited
when carrying out the role.
The interaction model expresses relationships and
dependencies between the various roles in a multi-
agent system. Interactions need to be captured and
represented in the analysis phase. Links between
roles are represented by the interaction model. It con-
sists of a set of protocol definitions, one for each type
of inter-role interaction. A protocol is a pattern of in-
teraction that has been formally defined and ab-
stracted away from any particular sequence of execu-
tion steps.
Design. The aim in Gaia’s design phase is to
transform the analysis models into a sufficiently low
level of abstraction so that traditional design tech-
niques including object-oriented techniques may be
applied in order to implement agents. The Gaia de-
sign process involves generating three models: The
agent model identifies the agent types that will make
up the system, and the agent instances that will be
instantiated from these types. The service model
identifies the main services that are required to carry
out the agents’ roles. Finally, the acquaintance
model documents the communication links between
the different agents.
2.2 Passi
The Process for Agent Societies Specification and
Implementation (PASSI) is a step-by-step require-
ment-to-code methodology (Costenio, and Potts,
2002). It consists of five models and twelve process
steps for building multi-agent systems and makes in-
tense use of the UML notation.
Analysis. The analysis is described in detail by
the system requirements model that focuses on the
HOW TO BUILD A MULTI-MULTI-AGENT SYSTEM: THE AGENT.ENTERPRISE APPROACH
365
requirements in terms of agency and purpose. UML
use-case diagrams are used to describe the agent
domain and help to identify the agents and their
roles within the system by analyzing their communi-
cation relationships.
Design. The design phase is characterized by the
Agent Society Model that consists of an ontology
description, a role description, and a protocol de-
scription. The ontology description visualizes two
ontologies: the domain ontology for the involved en-
tities and the communication ontology that models
the agents’ knowledge components and the inter-
agent communication. The advantage of this ap-
proach is the ability to model the agents’ knowledge
and the communication specification as two related
elements in the same way. The role description re-
fines the defined agents’ roles within the scope of
the agents’ lifecycles, collaboration, and communi-
cation. For formal communication protocol descrip-
tions AUML is recommended.
Implementation. The agent implementation
model defines the resulting system architecture by
developing a multi-agent and a single-agent struc-
ture definition. Furthermore a behavior definition us-
ing activity diagrams is used to describe the multi-
agent system.
All models have to be converted into a code
model. Following the implementation, the deploy-
ment model is used to describe the location of the
agents regarding the processing units and the migra-
tion and mobility constraints.
The strength of PASSI is the integration of known
object oriented design methods into multi-agent sys-
tem design.
2.3 Massive
MASSIVE stands for “Multi-Agent SystemS Itera-
tive View Engineering” and provides a framework
for the development of multi-agent systems (Lind,
2001). It combines new software development ap-
proaches and standard software engineering tech-
niques and applies them to multi-agent systems. The
MASSIVE method differs from other traditional
methodologies by being rather view-oriented than
process-oriented, which makes it pointless to sepa-
rate analysis and design.
Views are used to describe different aspects of the
complete design. The logical decomposition of a
system contains seven views: task, environment,
role, interaction, society, architecture and system.
The task view depicts the system from a functional
perspective whereas the environment view analyses
the accessibility of the environment from the system
and the developer perspective. The role view de-
scribes the role model for the agents by analyzing
the functional and physical relations within the sys-
tem. The interaction view models the interaction
necessary for solving a problem and the society view
takes a macro perspective for building up structured
collections of MAS entities. The resulting software
model for the MAS and the single agents are de-
scribed in the architectural and the system view.
The views are embedded in a stepwise refinement
of the process model, the so called iterative view en-
gineering. Due to this approach the model can deal
with an incomplete problem specification and is not
fixed for the entire project lifetime.
Another characterizing part of MASSIVE is the
experience factory that provides a conceptual
framework for enabling a systematic learning proc-
ess within an organization. This way it is possible to
improve the model according to the experience
gained in the development process.
3 THE AGENT.ENTERPRISE
APPROACH
The Agent.Enterprise initiative is a joint platform
within the priority research program of the Deutsche
Forschungsgemeinschaft to integrate recent research
results and to join forces in order to build up a net-
worked, agent-based application scenario for the
manufacturing domain (Frey et al., 2003). In order to
address the difficulties caused by the distributed
structure of the involved projects, we developed a
methodology based on aforementioned concepts of
agent-oriented software engineering. The following
subsections outline our scenario, the developed
methodology, and some underlying design decisions.
Scenario
The research groups participating in
Agent.Enterprise cover supply-chain-related aspects
of enterprise co-operation as well as intra-
organizational tasks of individual enterprises. In the
context of the MMAS scheduling functions are of-
fered by the DISPOWEB
*
project, shop floor pro-
duction planning and control are provided by the
KRASH
project, the IntaPS
project and the
*
Dispositive Supply-Web-Coordination, Available from:
http://www.dispoweb.de [Accessed 13.09.03]
Karlsruhe Robust Agent Shell, Available from:
http://www.ipd.uka.de/KRASH/ [Accessed 13.08.03]
Integrated Agent-based Process Planning and Produc-
tion Control, Available from: http://www.intaps.org
[Accessed 13.09.03]
ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING
366
FABMAS
§
project. Finally the ATT/SCC
**
project
provides proactive tracking and tracing services to
guarantee the reliability of supply chain processes in
the case of unforeseen disruptions.
Figure 1: Interaction in the integrated SCM architecture
A typical supply chain management cycle of dis-
tributed global planning of supply chain activities is
shown in Figure 1. After generating an initial plan of
orders and suborders comprising prices and dates of
delivery, software agents located at the different
supply chain partners carry out negotiations.
Thereby, they optimize the cost and the due dates of
deliveries (c).
These optimized delivery plans are used on the
intra-organizational level inside each enterprise (i.e.
in KRASH, IntaPS, and FABMAS) to plan the pro-
duction of goods on each stage of the supply chain
in detail. Three different MAS are concerned with
varying aspects of production planning (d). They
require the input from DISPOWEB agents and gen-
erate detailed plans for their production facilities.
These plans are the initial input for a controlling
system, which is developed in the ATT/SCC project.
The latter MAS monitors orders on every stage of
the supply chain using a distributed architecture in
order to proactively detect events that endanger the
planned fulfillment. In case of such an event, e.g. a
disruption in a production cycle, the ATT system in-
forms the related partner enterprises about the event
§
Agent-Based System for Production Control of Semi-
conductor Manufacturing Processes, Available from:
http://www.tu-ilmenau.de/fabmas/ [Accessed 13.09.03]
**
Agent-based Tracking and Tracing of Business Proc-
esses, Available from: http://www.wi2.uni-erlangen.de/
research/ATT/index-e.html [Accessed 13.09.03]
(e). This information can be used to trigger the re-
scheduling of the production steps on an enterprise
level (f) or, in case of major events, even in the re-
negotiation of the contracts on the inter-enterprise
level of the DISPOWEB system (g).
3.1 Agent.Enterprise Methodology
Unlike GAIA, PASSI, MASSIVE and other agent-
oriented development methods the Agent.Enterprise
methodology focuses on a distributed and weakly
coupled development process, while minimizing the
time required for face-to-face communication. Con-
sequently, the initial design period is comparatively
short and restricted to create the speech acts and in-
teraction protocol design.
The result of the analysis and design process are
consolidated in functionally restricted prototypes,
which constitute a test bed for the components of the
evolving MMAS. The projects substitute their proto-
types with gateway-agents to connect their applica-
tions to the common scenario requiring a process of
repeated cycles of redesign, implementation, and
tests. Figure 2 depicts our development approach,
while a detailed description can be found in the fol-
lowing subsections.
Analysis. To assign the participating projects to a
specific functionality within the supply chain, the
focus of the related research was taken as the major
criterion which roles each project would play. Yet,
the integration work starts solely with the Role
Definition and Assignment followed by the Use
Case Specification – the next step is to bring life to
the roles.
A first approximation of the Role Definition and
Assignment is made by performing a simple role-
playing technique. To simulate the exchange of in-
formation between the systems a member of the sci-
entific staff of each of the participating projects
takes over the role of her MAS, and writes down its
informational requirements. Then cards are handed
out. The sender writes down the contents of the mes-
sage as well as the receiver, therefore each card
represents a single act of communication. Starting
with the initiator of an order – the customer – the
whole supply chain is acted out until finally the last
card announces the delivery of the order from the
OEM to the customer. As a result of this role-
playing technique, the communication acts between
the projects’ MAS as well as the required informa-
tion are specified informally and can be formalized
into a Use Case Specification without requiring fur-
ther interaction between the participating projects.
HOW TO BUILD A MULTI-MULTI-AGENT SYSTEM: THE AGENT.ENTERPRISE APPROACH
367
Figure 2: The Agent.Enterprise development process
Speech Act Design. After defining the roles for
each participating MAS, it is necessary to ensure
that high-level communication between the systems
is possible. Due to the heterogeneity of knowledge
representation and semantics in the individual sys-
tems a language barrier for the communication ex-
ists. Consequently we introduce the so-called Gate-
way-Agent Concept, which is outlined in the next
subsection. This concept defines a virtual MAS
where the agents are scattered across a number of
agent platforms such that an ontology can specify
the semantics of conversations. While using onto-
logical expressions as a means of communication,
there are basically two methods to ensure that the
partners understand each other: Shared ontologies or
semantic mediation. As a first approach, we chose to
agree on a shared ontology for communication be-
tween gateway-agents. In our scenario of 5 commu-
nicating gateways the setup cost for implementing
semantic mediation would exceed the cost for agree-
ing on a shared ontology (Wache et al., 2001). Fu-
ture work will include the semantic mediation based
on a common terminology to further open the
Agent.Enterprise system to other MAS.
The task of ontological modeling is performed us-
ing the method described in (Noy, and McGuinness,
2001), which is supported by an ontology modeling
tool
††
. There is a number of tools, which support on-
tology-modeling, ranging from Protégé (Gennari et
al., 2002), OilEd (Bechhofer et al., 2001) to Web-
Onto (Domingue, 1998). Protégé has proven to be
best suited for the task of ontology modeling in our
context. The Protégé-plug-in Beangenerator (van
Aart et al., 2002) is able to generate code, which can
be used for the communication funded on the mod-
eled ontology. The code is deployed on the JADE
agent platform, a development framework employed
for several projects within the Agent.Enterprise
group.
A starting point for an ontology modeling is to
identify actions of the agents, which are requested
from a communication partner. They are directly de-
rived from assigned functionalities and specify tasks
to perform. Afterwards, ontological concepts defin-
ing the artifacts to be dealt with in agent actions can
be specified, e.g. the products to be manufactured.
After modeling all the details for the supply chain,
the concepts required for the supervision of all par-
ticipating MAS are designed.
Interaction-Protocol Design. The next step in
the overall design process is to define dynamics in
conversations, i.e. which interaction-protocols have
to be used for communication. The informal specifi-
cation of interactions resulting from the card role-
play is mapped to corresponding FIPA interaction
protocols if available. As a result, the behavior of
each gateway-agent for each MAS is specified as far
as communication between gateways is concerned.
The final step for integration is to ensure that the
communication between the gateway-agent and the
underlying MAS works.
(Distributed) Implementation. Based on the dis-
tributed structure of our research program, the de-
velopment process takes into account comparatively
long periods of independent development. Inspired
by the concepts of Extreme Programming (Beck,
1999), the development process starts by the imple-
mentation of functionally restricted prototypes exe-
cuting a simplified test case. These prototypes serve
two purposes: On the one hand a consolidation of
speech acts and interaction protocols is enforced. On
the other hand, a test bed for autonomous develop-
††
An overview of methodologies to develop ontologies
has been given in Fernández-López, and Gómez-Pérez
(2002).
ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING
368
ment emerges. The implementation of these proto-
types is closely bound to test-sessions. The hereby
gained experience is used for improvement and re-
finement of speech acts and/or interaction protocol
design. The outcome of this work is a set of test
modules and an exchange of experience within the
covering research program. As an inevitable restric-
tion of our approach, the simplification of the central
projects’ functions in the ‘Dummy Gateway Imple-
mentation’ could result in a setting where some as-
pects of the interaction protocols could not be suffi-
ciently tested.
Subsequent to the completion of the prototypes,
each project integrates its fully functional applica-
tion into the test bed. Obviously each project some-
times has to debug prototypes of other projects. Re-
sulting from clearly defined responsibilities for each
prototype, explicit phases for consolidation are not
intended.
3.2 The Gateway-Agent Concept
Integration of complex systems needs agreements of
technical nature in order to avoid a time-consuming
struggle with implementation details. For the
Agent.Enterprise approach two central design deci-
sions are subsumed in the Gateway-Agent Concept,
which is illustrated in Figure 3.
Firstly, the agreement upon the use of FIPA-
compliant platforms avoids many of the communica-
tion-related obstacles and allows for concentrating
on domain aspects. The second decision is that every
individual MAS to be integrated should be repre-
sented by a single agent in the resulting MMAS.
FIPA-compliant MAS-Platform
(e.g. FIPA-OS)
FIPA-compliant MAS-Platform
(e.g. JADE)
MAS B
(e.g. SCM)
MAS A
(e.g. PPC)
GW-
Agent
A
GW-
Agent
B
interaction
Figure 3: The Gateway-Agent Concept
Thus, the interacting gateway-agents build up a
virtual MAS. Together, these decisions can be seen
as a specific MMAS architecture, which subsumes
aspects of various well-known software patterns
(Gamma et al., 1995):
- Façade pattern: The gateway-agents provide a
unified interface to their MAS as a subsystem,
comprising different roles and their respective
functionality.
- Wrapper pattern (also called Adaptor): The
gateway agents translate between internal for-
mats and behavior of their corresponding MAS
and the common representation in the virtual
MAS.
- Bridge pattern: The different types of the gate-
way agents provide abstract interfaces decoup-
led from the implementation of their MAS. In
the next section we give an example of three
MAS that play the role of a supplier in a supply
chain scenario. In the virtual MAS they all are
represented by the same type of gateway agent
while their implementation is completely differ-
ent and independent.
There are several advantages of the Gateway-
Agent Concept, e.g., developers can focus their ef-
fort on a single agent and only the gateway-agents
must be available during operation. Also, there is no
restriction in the centralization of the different roles
of the MAS into a single gateway-agent, as the dif-
ferent functionalities of this agent can be redirected
to several other agents in the MAS it represents.
4 BENEFITS OF THE
AGENT.ENTERPRISE
METHODOLOGY
In spite of the many differences of the presented
methods, e.g. in respect to scope, representation
format, development process, and semantic founda-
tion, they share some central concepts. Agents pro-
vide functionality by taking over certain tasks asso-
ciated with a role. They interact in order to build up
societies and have a domain model of the environ-
ment in which they are situated.
The presence of the concepts and their corre-
sponding models in the described development
methods on one hand raises the question if and how
they must be addressed by a design process, which
aims at the integration of several existing independ-
ent MAS to an MMAS. On the other hand, such an
integration method needs not to cover all aspects of
agent-oriented development. It has to concentrate on
the requirements from the integration of different al-
ready existing systems.
HOW TO BUILD A MULTI-MULTI-AGENT SYSTEM: THE AGENT.ENTERPRISE APPROACH
369
Table 1. MAS-specific concepts and aspects covered by the discussed methods
Gaia PASSI MASSIVE
Agent.Enterprise
Functionality
- (input) Domain Description Task View - (input)
Roles / Agents
Role Model
/ Agent Model
Agent and Roles Iden-
tification
/ Role Description
Role View Role Identification
and Description
Tasks
Role M. (activities),
Services Model
Task Specification Task View - (input)
Interaction
Interactions Model
/ Services Model
Protocols
Description
Interaction View Protocols
Description
Domain
Knowledge
- Ontology - Shared + Individual
Ontologies
Macro Level
- (not yet) (Role Description) Society View (Role Description)
Environment
Role M. (rights, perm.
and responsibilities)
- Environment View -
Architecture /
Implementation
- Agent Implementation
and Code M.
Architectural View Gateway Model
Deployment
Acquaintance Model Deployment Model System View (not yet)
Although our Agent.Enterprise approach is com-
parable to fully-fledged development methods, it
rather has to be seen as an addition to such methods.
For example, the functionality of the resulting
MMAS derives from existing MAS instead of being
newly designed. Hence certain aspects of the differ-
ent approaches can not be compared (cp. Table 1). In
the following we emphasize three aspects where
conventional methodologies could not be applied.
Architecture. The Gateway-Agent Concept calls
for the assignment of a specific functionality to the
corresponding gateway-agent. One could say that the
gateway-agents give a macro level view of the
MMAS as representatives of their specific MAS.
The distinction between functionality and role can
be unattended in this context, as a specific function-
ality is carried out by individual MAS and therefore
its gateway-agent is required to play the role associ-
ated with the functionality.
Domain Knowledge. In most development meth-
ods, the modeling of the domain knowledge does not
play a central role. Often an implicit common do-
main model is assumed when a (closed) MAS is de-
veloped from scratch. This assumption does not hold
for the integration of existing MAS. Thus, the design
of a domain model, which is common to all gate
way-agents, and suitable at least for communication
purposes becomes a major development step.
The same applies to the interaction design: it has
to consider the dynamic aspects of the functionality
provision and leads to a number of communication
protocols. A basic experience of Agent.Enterprise is
the necessity of a bottom-up approach for the design
of interactions caused by the established functional-
ity of the participating MAS. The interaction design
comprises three elements which are similar to the
PASSI method (but less formalized): informal speci-
fication, mapping to FIPA-interaction protocols
(AUML), and ontology-based content specification
(e.g. Protégé).
All other mentioned aspects such as accessibility
of the environment and deployment can be left to the
individual projects as no overall design guideline is
necessary for integration. The result of the applica-
tion of our methodology is the fully functional
Agent.Enterprise MMAS.
5 CONCLUSION
This paper introduces a new methodology for cou-
pling MAS using methods from agent-oriented de-
sign. The existence of sufficiently specified concepts
applicable for most steps of our prospected method-
ology made little conceptual effort necessary.
Our approach addresses special needs, e.g. deal-
ing with individual ontologies that arise from the na-
ture of MAS integration. Special attention is given to
agreements on the technical system architecture. The
introduced technical concept, called Gateway-
Agents, accelerates the implementation by enforcing
those necessary agreements on technical standards.
Furthermore we could replace an often inflexible
standardization of interfaces by tool-supported ver-
bal agreements. Moreover, the Agent.Enterprise
methodology is designed to fit the need for distrib-
uted development.
ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING
370
The resulting MMAS called Agent.Enterprise
covers services in the range of supply chain schedul-
ing, shop floor production planning and control, and
proactive tracking and tracing services. By building
this large-scale adaptable MMAS we showed the
applicability of our methodology. The provided reli-
ability of overall supply chain processes also sup-
ports our assumption that MMAS will turn out to be
a valuable extension to MAS esp. in the context of
Supply Chain Management.
ACKNOWLEDGEMENTS
This work as well as the participating projects is
funded by the Deutsche Forschungsgemeinschaft
(DFG) within the German priority research program
1083 “Intelligent Agents in Real-World Business
Applications” (refer to http://www.realagents.org for
further information).
We also like to thank Daniel Pfeifer, Michael H.
Schwind and Ingo J. Timm for their contributions.
REFERENCES
van Aart, C. J., Pels, R.F., Giovanni, C. & Bergenti, F.
(2002) ‘Creating and Using Ontologies in Agent
Communication’, Proc. of the Workshop on Ontolo-
gies in Agent Systems, 1st International Joint Confer-
ence on Autonomous Agents and Multi-Agent Systems
(AAMAS2002), Bologna, Italy.
Bechhofer, S., Horrocks, I., Goble, C. & Stevens, R.
(2001) ‘
OilEd: a Reason-able Ontology Editor for the
Semantic Web
‘, Proc. of Joint German/Austrian con-
ference on Artificial Intelligence (
KI2001), Springer-
Verlag.
Bellifemine, F., Poggi, A. & Rimassa, G. (2001) ‘JADE: a
FIPA2000 Compliant Agent Development Environ-
ment‘, Proc. of the Fifth International Conference on
Autonomous Agents, ACM Press.
Cosentino, M. & Potts, C. (2002) ‘A CASE Tool Sup-
ported Methodology for the Design of Multi-Agent
Systems‘, Proc. of the International Conference on
Software Engineering Research and Practice, CSREA
Press.
Domingue, J. (1998) ‘
Tadzebao and WebOnto: Discuss-
ing, Browsing, and Editing Ontologies on the Web’,
Proc. of the 11th Workshop on Knowledge Acquisition
for Knowledge-based Systems, Banff, Canada.
Fernández-López, M. & Gómez-Pérez, A. (2002) ‘Over-
view and Analysis of methodologies for building on-
tologies’ The Knowledge Engineering Review (KER),
vol. 17 [2]. pp: 129-156.
Frey, D., Stockheim, T., Woelk, P.-O. & Zimmermann, R.
(2003) ‘Integrated Multi-agent-based Supply Chain
Management’, Proc. of the 1st International Workshop
on Agent-based Computing for Enterprise Collabora-
tion, IEEE Computer Society Press.
Gamma, E., Helm, R., Johnson, R. & Vlissides, J. (1995)
Design Patterns: Elements of Reusable Object-
Oriented Software’, Addison-Wesley, Reading, Mass.
Gennari, J., Musen, M. A., Fergerson, R. W., Grosso, W.
E., Crubézy, M., Eriksson, H., Noy, N. F. & Tu, S. W.
(2002) ‘The Evolution of Protégé: An Environment for
Knowledge-Based Systems Development‘, Medical
Informatics Technical Report SMI-2002-0943, Stan-
ford.
Lind, J. (2001) ‘Iterative Software Engineering for Multi-
agent Systems: the MASSIVE Method‘, Lecture Notes
in Computer Science and Lecture Notes in Artificial
Intelligence, Vol. 1994, Springer-Verlag.
Noy, N.F. & McGuinness, D.L. (2001) ‘Ontology Devel-
opment 101: A Guide to Creating Your First Ontol-
ogy‘. Stanford Knowledge Systems Laboratory Tech-
nical Report KSL-01-05 and Stanford Medical
Informatics Technical Report SMI-2001-0880, Stan-
ford, CA.
Odell, J., Van Dyke Parunak, H. & Bauer, B. (2001) ‘Rep-
resenting Agent Interaction Protocols in UML‘, Proc.
of the First International Workshop on Agent-oriented
Software Engineering, Springer-Verlag.
Special Interest Group on Manufacturing Logistics of the
German priority research program
1083 “Intelligent
Agents in Real-World Business Applications”
Wache, H., Vögele, T., Visser, U., Stuckenschmidt, H.,
Schuster, G., Neumann N. & Hübner, S. (2001) ‘On-
tology-Based Integration of Information - A Survey of
Existing Approaches’, Gómez-Pérez, A., Gruninger,
M., Stuckenschmidt, S. & Uschold, M. (eds.), Proc. of
the IJCAI-Workshop Ontologies and Information
Sharing, Seattle, WA, pp. 108-117.
Wood, M.F. & DeLoach, S.A. (2001) ‘An Overview of
the Multiagent System Engineering Methodology’,
Proc. of the First International Workshop on Agent-
oriented Software Engineering, Springer-Verlag.
Wooldridge, M., Jennings, N. R. & Kinny, D. (2000) ‘The
Gaia Methodology for Agent-Oriented Analysis and
Design‘, Journal of Autonomous Agents and Multi-
Agent Systems, vol. 3, no. 3, pp. 285–312.
HOW TO BUILD A MULTI-MULTI-AGENT SYSTEM: THE AGENT.ENTERPRISE APPROACH
371