Analyzing dynamic behavior of multiagents' task allocation based on subjective cooperative relations

In multiagent systems in general, the behavior of the individual agent generally affects the behavior of the whole system. Consequently, in order to improve the efficiency of the whole system, it is crucial to know how the operation of the individual agent affects the whole system and to design the agents on the basis of that knowledge. This paper considers a multiagent system in which the technique proposed by the authors is applied to task allocation, and presents a basic analysis of the effect of the individual agent on the behavior of the whole system. More precisely, we consider an environment in which the procedure for updating information concerning the reliance relation to other agents, or the procedure for transfer of information concerning the power of other agents, and so on, is modified, or agents in the system are dynamically generated or deleted, and we analyze the efficiency of task processing and the amount of message traffic among the agents. Experiments are performed by defining several kinds of operations of the agents, and generating or deleting the agents while the system is operating. Based on the results of these investigations, basic guidelines for multiagent systems are presented.