A Hybrid Architecture for Hierarchical Agents

This paper presents a hybrid multi-agent architecture called MAGIQUE 1. The aim of this architecture is to take advantage of both blackboard systems and autonomous agents. Our model contains several hierarchical control levels. It allows users modeling complex applications that involve interactions of sets of agents. For example in a multi-expert system, we can have a level for each logic used for deduction specialized in the communication between agents of a group. The paper describes the model and specify functionalities of two kinds of agents: specialists, and supervisors. Specialists hold expertise knowledge and are specialized in solving speciic tasks, whereas supervisors coordinate interactions between a group of agents which are described by their mental attitudes (beliefs and goals). At the end of this paper, we shows also an example of application we are realized with MAGIQUE. We conclude by indicating the future application areas, particularly the robotics.