A Formal Environment Model for Multi-Agent Systems - Formal Methods: Foundations and Applications - page 64

A Formal Environment Model for Multi-Agent
Systems

Paulo Salem da Silva and Ana C.V. de Melo

University of São Paulo
Department of Computer Science
São Paulo, Brazil
[e-mail] , [e-mail]

Abstract. Multi-agent systems are employed to model complex systems
which can be decomposed into several interacting pieces called agents. In
such systems, agents exist, evolve and interact within an environment.
In this paper we present a model for the specification of such environ-
ments. This Environment Model for Multi-Agent Systems (EMMAS), as
we call it, defines both structural and dynamic aspects of environments.
Structurally, EMMAS connects agents by a social network, in which the
link between agents is specified as the capability that one agent has to
act upon another. Dynamically, EMMAS provides operations that can
be composed together in order to create a number of different environ-
mental situations and to respond appropriately to agents' actions. These
features are founded on a mathematical model that we provide and that
defines rigorously what constitutes an environment. Formality is achieved
by employing the π -calculus process algebra in order to give the seman-
tics of this model. This allows, in particular, a simple characterization
of the evolution of the environment structure. Moreover, owing to this
formal semantics, it is possible to perform formal analyses on environ-
ments thus described. For the sake of illustration, a concrete example of
environment specification using EMMAS is also given.

1 Introduction

Multi-agent systems (MAS) [13] can be used to model complex systems in which
the entities to be studied can be decomposed into several interacting pieces called
agents . Human societies, computer networks, neural tissue and cell biology are
examples of systems that can be seen from this perspective. Given a MAS,
one technique often employed to study it is simulation [2]. That is, one may
implement the several agents of interest, compose them into a MAS, and then
run simulations in order to analyse their dynamic behavior. In such works, the
analysis method of choice is usually the collection or optimization of statistics
over several runs (e.g., the mean value of a numeric variable over time). Examples
of this approach include platforms such as Swarm [5], MASON [3] and Repast
[6]. There are, however, other possibilities for analysing such simulations. The
crucial insight here is that simulations can be seen as incomplete explorations of
state-spaces, and thus can be subject to some kinds of formal analyses.