Learning of Task Allocation Method Based on Reorganization of Agent Networks in Known and Unknown Environments

Abstract

We propose a team formation method that integrates the estimating of the resources of neighboring agents in a tree-structured agent network in order to allocate tasks to the agents that have sufficient capabilities for doing tasks. A task for providing the required service in a distributed environment is often achieved by a number of subtasks that are dynamically constructed on demand in a bottom-up manner and then done by the team of appropriate agents. A number of studies were conducted on efficient team formation for quality services. However, most of them assume that resources in other agents are known, and this assumption is not adequate in real world applications. The contribution of this paper is threefold. First, we extend the conventional method by combining the learning of task allocation and the reorganization of agent networks. In particular, we introduce the elimination of links as well as the generation of links in the reorganization. Second, we revise the learning method so as to use only information available locally. Finally, we omitt the assumption that all resource information in other agents is given in advance. Instead, we extend the task allocation method by combining it with the resource estimation of neighboring agents. We experimentally show that this extension can considerably improve the efficiency of team formation compared with the conventional method even though it does not require knowledge of resources in other agents. We also show that it can make the agent network adaptive to environmental changes.