PAGODA (Policy And GOal Based Distributed Autonomy) is a modular architecture for specifying and prototyping autonomous systems. A PAGODA node (agent) interacts with its environment by sensing and affecting, driven by goals to achieve and constrained by policies. A PAGODA system is a collection of PAGODA nodes cooperating to achieve some mutual goal. This paper describes a specification of PAGODA using the Russian Dolls model of policy-based coordination. In PAGODA there are two forms of coordination: local and global. Local coordination is used to compose the components of a PAGODA node. The local coordinator is concerned with ensuring component level synchronization constraints, cross component message ordering constraints, routing of notifications, and interaction with the external world. The global coordinator is concerned with dissemination of information, negotiation of responsibilities, and synchronization of activities. Requirements for a PAGODA node coordinator are given and an example set of policies is specified. Principles for showing that the policies satisfy the requirements are discussed as a first step toward a logic of policy-based coordination. Development of a distributed coordinator is the subject of ongoing work. Some challenges and possible solutions are discussed.
Keywords: coordination, composition, distributed object reflection, goal, autonomy,policy