Task Decomposition and Delegation Algorithms for Coordinating Unstructured Multi Agent Systems

Abstract

This paper describes the initial PhD research work for developing an innovative process for agent coordination in unstructured distributed environments. In this proposal, an agent that receives a request will use an AI-based planning algorithm to decompose the received request into several subtasks; it will handle those for which it has know-how and available resources; and it will use decentralized networks computing methods to delegate the rest of the request to other agents that can possibly handle it. These, in turn, will recursively apply the same process until the initial request is completely solved or until some specified termination condition holds. The work contributions to the state-of-the-art are: the creation of an AI-based planning algorithm capable of partial task decomposition with only partial knowledge of available planning operators; and the creation of a coordination infrastructure that enables agents to delegate subtasks to other agents in totally unstructured decentralized environments