Page 136 -

AGENT-ORIENTED METHODS AND METHOD ENGINEERING 121
Although influenced by OMT, MAS-CommonKADS (Iglesias et al., 1998) also has strong AI/
knowledge engineering influences (CommonKADS: Schreiber et al., 1994). It is an agent-oriented
methodology that supports the development of MAS from the conceptualization phase through
to a detailed design that can be directly implemented. The main modeling concepts in MAS-
CommonKADS are agent, knowledge, organization, and coordination.
Being based to some degree on ideas in the OO Fusion methodology, itself based in part on
OMT, Gaia (Wooldridge, Jennings, and Kinny, 2000) views the process of multi-agent system
(MAS) development as a process of organizational design, where the MAS is modeled as an or-
ganized society with agents playing different roles. The methodology allows a developer to move
systematically from a statement of requirements to a design detailed enough to be implemented
directly. It supports both macro (societal) and micro (agent) aspects of MAS design, and is also
neutral to both application domain and agent architecture. The newest version of Gaia (Zambo-
nelli, Jennings, and Wooldridge, 2003) extends the original version with various organizational
abstractions, enabling it to be used for the design of an open MAS (which was not achievable
previously). Another AO methodology said to extend Gaia is Societies in Open and Distributed
Agent spaces (SODA; Omicini, 2000). As its name implies, it focuses on societal descriptions
of multi-agent systems, especially Internet-based applications. SODA concentrates on interagent
issues and leaves the developer to choose his/her own internal agent model. Also initially influ-
enced strongly by Gaia is Role-Oriented Analysis and Design for Multi-Agent Programming
(ROADMAP; Juan, Pearce, and Sterling, 2002; Juan and Sterling, 2003). It introduces into Gaia
UML-style use cases for requirements gathering, explicit models of the agent environment, and
agent knowledge together with an interaction model based on AUML (Odell, Van Dyke Parunak,
and Bauer, 2000). ROADMAP draws a clear distinction between analysis and design with a focus
on roles and goals (as do many other AO methodologies).
A group of AO methods have been influenced by RUP (Kruchten, 1999) rather than the older
OMT. These include MESSAGE, INGENIAS, Adelfe, and AOR/RAP. MESSAGE (e.g., Garijo,
Gomez-Sanz, and Fuentes, 2005) arose from the needs of the telecommunications industry. It is
said to extend UML to cover analysis and design considerations, adding agent-specific concepts
to describe organizations, roles, goals, and tasks. INGENIAS (Pavón, Gomez-Sanz, and Fuentes,
2005) builds on the ideas of MESSAGE, focusing on five views: organization, agent, goals/tasks,
interactions, and environment. These viewpoints are then complemented by the use of extensions
of OO notations such as UML. The process elements in INGENIAS are based on those in the
UDP (Jacobson, Booch, and Rumbaugh, 1999). A set of interrelated activities (approximately
100) are then defined, which assist the developer in creating the final MAS specification. Also
using RUP as a basic input, RAP (Taveter and Wagner, 2005) uses the AOR notation of Wagner
(2003). Its focus is business processes that emerge as social interactions from the behavior of the
participating agents.
Although also using many RUP-like ideas, Adelfe (e.g., Bernon et al., 2002, 2005; Piquemal-
Baluard et al., 1996) is aimed at a very different kind of AO system than most other AO meth-
odologies. Adelfe is primarily intended to be used for the development of adaptive multi-agent
software applications. It tailors RUP and introduces its phases as either WorkDefinitions (WDi),
Activities (Aj) or Steps (Sk) following the vocabulary of the OMG’s Software Processing Engi-
neering Metamodel (SPEM) (OMG, 2002), which has been used to underpin Adelfe.
There are a number of AO methods that offer only a passing acknowledgment to OO. These
include PASSI and Prometheus. PASSI (Process for Agent Societies Specification and Implementa-
tion) (Burrafato and Cossentino, 2002; Cossentino, 2005) offers a step-by-step requirement-to-code
process for the development of a MAS (Figure 8.1), integrating models and concepts from both