190 likes | 377 Views
Unifying MAS Meta-Models ADELFE, Gaia & PASSI. Carole Bernon, Massimo Cossentino, Marie-Pierre Gleizes, Paola Turci and Franco Zambonelli. Agent-Oriented vs. Object-Oriented. Design a system Instantiate a meta-model OO context - design rely on a common denominator
E N D
Unifying MAS Meta-ModelsADELFE, Gaia & PASSI Carole Bernon, Massimo Cossentino, Marie-Pierre Gleizes, Paola Turci and Franco Zambonelli
Agent-Oriented vs. Object-Oriented • Design a system Instantiate a meta-model • OO context - design rely on a common denominator • Universally accepted concept of object and related meta‑model of object-oriented systems • AO context - to date, no common denominator • Each methodology has its own concepts and system structure In the agent world the meta-model is the critical element when applying the “method engineering” paradigm TFG AOSE, Rome, July 2004
Reasons to Unify • Meta-model = means to unify concepts • Unique meta-model, a way of unifying the different concepts • Helping designer to build a meta-model depending on the system he needs • Choice of useful elements • From the “omni-comprehensive” MAS meta-model • Composition of process • “Method fragment” selection phase • Model coherence checking by CASE tools • Unification eases model transformation TFG AOSE, Rome, July 2004
cognitive and behavioural representation MAS Meta-Model for ADELFE no predefined organization • local goal • cooperative • detect and remove NCS TFG AOSE, Rome, July 2004
Gaia Methodology • First version • Designed to handle small-scale, closed agent-based systems • Modelled agents, roles, interactions • Missed in modelling explicitly the social aspects of a MAS • Official extension of Gaia • Thought for open agent systems • Focused on the social organization of the system TFG AOSE, Rome, July 2004
MAS Meta-Model for Gaia TFG AOSE, Rome, July 2004
MAS Meta-Model for PASSI TFG AOSE, Rome, July 2004
Towards a Unifying MAS Meta-Model • In order to define a unifying meta-model we will consider the following specific aspects: • Agent structure • Agent interactions • Agent society and organizational structure • Agent implementation TFG AOSE, Rome, July 2004
Agent Structure • ADELFE: cooperative agents • Gaia and PASSI: composition of roles TFG AOSE, Rome, July 2004
Agent Structure - Considerations • None imposes a specific classical agent architecture • Seen as some kind of low level architectures that can be adopted during MAS implementation • None explicitly deals with goals and plans • ADELFE • Notion of goal is only used to determine skills • Plans are built at run-time by the system • Gaia • The concept of “goal” is implicit in roles • Plans play no explicit role • PASSI • Goals are considered as non functional requirements • Plans are modelled as algorithms (activity diagrams) TFG AOSE, Rome, July 2004
Agent Interactions • No relevant differences TFG AOSE, Rome, July 2004
Agent Interactions • No relevant differences • ADELFE: agents can communicate indirectly using environment • Gaia: communication mediated by the environment seen as a side effect TFG AOSE, Rome, July 2004
Agent Interactions • No relevant differences • ADELFE: agents can communicate indirectly using environment • Gaia: communication mediated by the environment seen as a side effect • ADELFE: ontologies have not to be modelled; agents are able to adapt to the environment and other agents TFG AOSE, Rome, July 2004
Agent Society and Organizational Structure • ADELFE: no predefined organization TFG AOSE, Rome, July 2004
Agent Society and Organizational Structure • ADELFE: no predefined organization • Gaia: organization = primary abstraction • PASSI: services and scenarios TFG AOSE, Rome, July 2004
Agent Implementation • ADELFE • The problem of the system implementation has not been treated yet; no platform has been imposed • Gaia • Totally abstract from implementation • PASSI • FIPA compliant systems • Direct map among the most important elements of the model and their implementation TFG AOSE, Rome, July 2004
Proposal for a Unifying Meta-model TFG AOSE, Rome, July 2004
Conclusion • Societies with or without predefined organizations (open systems) • Generic agent (cooperative or not) • Follows organizational or cooperation rules • Possibility of identifying three domains • Improvements: an agent has a representation of the environment expressed via an ontological model • Step towards a unique meta-model • A unifying framework for the systems produced with different approaches TFG AOSE, Rome, July 2004
Open Questions • Can a unique meta-model exist? • What description level has to be reached? • How may designers choose meta-model elements? • What kind of tools may ease their work? TFG AOSE, Rome, July 2004