Incrementally Refined Acquaintance Model for Distributed Planning in a Semi-Trusted Environment

This paper analyzes the distributed planning architecture proposed by Durfee (Durfee 1999) and discusses the properties of the suggested architecture in a peerto-peer, dynamic and semi-trusted environment. The I-Globe integrated architecture linking classical planning algorithms with multi-agent technologies is presented (namely using the Hierarchical Task Network style I-X technology and A-globe multi-agent system). As a key technical contribution of the paper we present a specific interaction protocol based on incremental improvement of the social knowledge of agents, the Incrementally Refined Acquaintance Model (IRAM), and provide justification for its deployment in a specific distributed planning scenario.