Méthodologie de développement de systèmes multi-agents adaptatifs et conception de logiciels à fonctionnalité émergente

摘要

Environments within which applications are embedded are growing in complexity and dynamicity, considering the large number and the diversity of the takeholders. Functions of such systems become more and more difficult to define, and their specifications are often incomplete, even if their components are easily identifiable and specifiable. Without new design and modeling methods, managing such project will become too constraining, long and costly to cope with.We propose to use cooperative self-organising adaptive multi-agent systems (AMAS) to tackle these design problems. The functionality of such systems emerges from coopera- tive interactions between agents. Nevertheless, developing using AMAS is still an ad-hoc process and reduced to a small group of users. Several applications have been designed by using AMAS, but it has never been executed by novices and non AMAS experts. To answer to this lack of visibility and openess, the ADELFE project – for Atelier de DEveloppement de Logiciels à Fonctionnalité Emergente or Toolkit for developing applications with emergent functionalities – proposes to develop a methodology based on these emergence oriented principles. This methodology is defined in three points : process, notations and tools. The ADELFE process is based on the Rational Unified Process and extends or adds some agent specific activities. Notations are extensions of UML and A-UML. Tools have been developed or extended in to support notations, with OpenTool, and following the process is eased by using AdelfeToolkit.The relevance of this approach has been confronted to the development of experimental applications. Some results from a dynamic timetable solver, ETTO, and from a multi-robot transportation task are presented and analysed.