A Process Based on the Model-Driven Architecture to Enable the Definition of Platform-Independent Simulation Models - Simulation and Modeling Methodologies, Technologies and Application - page 109

The remainder of this paper is organized as follows: available ABMS languages,
methodologies and tools are briefly discussed along with the main drawbacks which
still hinder their wider adoption (Section 2), the proposed MDA-based process for
ABMS (MDA4ABMS) is presented (Section 3) and then exemplified (Section 4) with
reference to a popular problem (the Demographic Prisoner's Dilemma ) able to
represent several social and economic scenarios. Finally, conclusions are drawn and
future works delineated.

2 ABMS: Languages, Methodologies and Tools

Several approaches have been proposed to support the definition of agent-based mod-
els and/or their implementation for specific simulation platforms; in the following,
these approaches, grouped on the basis of the main features they provide, are briefly
discussed and their main drawbacks, which still hinder their wider adoption, are high-
lighted.
1. Agent-based Modeling and Simulation Platforms. ABMS platforms, which also
provide a visual editor for defining simulation models and, in particular, for specify-
ing agent behaviours, as well as semi-automatic code generation capabilities, are cur-
rently available, e.g. Repast for Python Scripting (RepastPy) [11], Repast Simphony
(Repast S) [29], the Multi-Agent Simulation Suite (MASS) [19], Ascape [35], SeSAm
[25], and Escape [3].
Although the existing ABMS tools attempt to offer comfortable modeling and si-
mulation environments, their exploitation is comfortable only when used for simple
models. In fact, to model complex systems where basic behavior templates provided
by the tools must be extended, significant programming skills are essential. Moreover,
as these tools do not refer to any specific ABMS process, their use is mainly based on
the extension and refinement of the examples and case studies provided, thus limiting
such platform-dependent models to lower levels of abstraction and flexibility. Finally,
the agent models adopted are often purely reactive and do not take into account orga-
nizational issues.
2. Agent-based Modeling Languages. Agent modeling languages, mainly coming
from the Agent-Oriented Software Engineering (AOSE) domain, can be exploited for
a clear, high level and often semantically well-founded definition of ABMS models;
some of the wider adopted proposals are the Agent-Object-Relationship (AOR) Mod-
eling [42], the Agent UML (AUML) [5], the Agent Modeling Language (AML) [10]
and the Multi-Agent Modelling Language (MAML) [20].
These languages, which do not refer to a specific modeling process, are high-level
languages based on graphical and, in some cases, easily customizable notations. Their
capabilities make them more suitable as languages for depicting models than as pro-
gramming languages. Moreover, compared to the models offered by agent-based si-
mulation toolkits, the agent models expressed by these languages are richer, both at
micro (agent) and macro (organization) levels. However, the definition of these agent
models often requires advanced modeling skills and the transition from the produced
design models to specific operational models must be often manually performed; this