MAGE: Multi-Agent Game Environment

Visara Urovi

Research output: ThesisDoctoral Thesis

236 Downloads (Pure)


We study the use of games as a metaphor for building social interaction
in norm-governed multi-agent systems. As part of our research we propose
MAGE (Multi-Agent Game Environment) as a logic-based framework that
represents complex agent interactions as games. MAGE seeks to (a) reuse
existing computational techniques for defining event-based normative sys-
tem and (b) complement these techniques with a coordination component to
support complex interactions.
A game in MAGE is defined by a state, a set of normative rules describing
the valid moves at different states and a set of effect rules describing how the
state evolves as a result of a move taking place. Given a specification of the
normative rules, in the implementation of a game, we use game containers as
components that mediate the moves of players by checking their compliance
with the rules of the game and by maintaining the state of the game.
The reuse part of MAGE relates physical actions that happen in an agent
environment to valid moves of a game representing the social environment of
an application. MAGE allows to model complex interactions from simpler
atomic sub-games. In this context, we investigate how coordination patterns
can be used to dynamically play more than one game in parallel, change the
status of games or choose amongst games. For this purpose, we examine
how to define compound games from atomic sub-games. Compound games
are build by describing the conditions and the patterns that activate a sub-
game and the coordination mechanisms of MAGE ensure that sub-games are
activated according to how interactions are specified to evolve at run-time.
To illustrate the MAGE approach, we discuss how to use the framework
to specify the social interaction in two different scenarios: (i) Open-Packet-
World - a simple simulation where agents compete to collect and deliver
packets in a grid and (ii) an earth-observation application - where agents
represent services, both for clients and providers, and negotiate the provision
of these services by combining argumentation and communication protocols.
We also use the Open-Packet-World scenario to evaluate the effectiveness of
the framework. We show that we can effectively support at run-time a large-
scale multi-agent systems regulated by norms. We conclude the dissertation
by summarising our contributions and identifying areas for future work.
Original languageEnglish
Awarding Institution
  • Royal Holloway, University of London
  • Stathis, Kostas, Supervisor
Award date1 May 2011
Publication statusUnpublished - 23 Mar 2011

Cite this