Techniques for Multi-agent System Reorganization

摘要

A multi-agent system which operates in an open environment must be able to react to unpredictable events. These events lead, at the system level, to possible system's failures and, inside the system, to agents' failures. Each agent performs several roles which could be unfulfilled in the system in case of agents' failures. To overcome these failure situations, agents could have their interactions and/or roles change during the multi-agent system execution. Doing so, we can prevent from system incoherence and possible deadlocks. Hence, we propose in this paper two techniques such that the first techniques allows to operate changes in agents' interactions and the second technique allows to operate changes in agents' roles in order to build adaptive multi-agent systems. We will illustrate our techniques by applying them to a case study: a timetable design.