A Policy Based Framework for Software Agents

摘要

In this paper we provide a policy based framework for managing and modelling agent social interaction. We specify and implement agent conversation patterns and protocols using an agent policy language. These patterns classify agent interaction, using a blend of communication protocols, security and management policies as well as Object Oriented software design principles. All services, including management services, interact with their agent clients using the same conversation patterns as defined in this paper. Any conversation begins with three steps: requesting a service, negotiating a conversation pattern, and establishing pattern roles. Role theory and reusable policy specifications regulate the way agents participate, providing a rich source of information for conversation management. Further, role theory is used to define agent roles and relationships, which are then described using the Agent Policy Language in terms of policies (i.e. obligations and authorisations). Policy based agents have the ability to interpret policies and assume roles for a certain interaction pattern dynamically. The methodology promotes platform independence and fits the needs of a modular, distributed environment; enabling services to use the powerful 'plug-and-play' concept. Co-operation patterns being more complex are built upon conversation patterns, but also describe the 'social' relationships between agents based on beliefs, desires and intentions.