Anurag.Acharya@CENTRO.SOAR.CS.CMU.EDU (03/12/88)
AI SEMINAR TOPIC: The Partial Global Planning Approach to Coordinating Distributed Problem Solvers SPEAKER: Edmund H. Durfee Department of Computer & Information Science Lederle Graduate Research Center University of Massachusetts at Amherst Amherst, Massachusetts 01003 (413) 545-1349 WHEN: Tuesday, March 15, 1988 3:30pm WHERE: Wean Hall 5409 ABSTRACT As distributed computing is used in applications where the distributed tasks are highly interrelated and change dynamically, coordination becomes increasingly important and difficult. Distributed artificial intelligence (DAI) applications often have these characteristics. In distributed problem solving networks, for example, individual nodes solve interacting subproblems of larger network problems. Network problems may change over time, so effective network problem solving depends on nodes coordinating their local actions and planning their interactions to cooperate as a coherent team. We introduce partial global planning as a new approach to coordination. Whereas previous DAI approaches, such as contracting or multi-agent planning, are specialized for particular situations, our new partial global planning approach provides a unified and versatile framework for dynamically coordinating independent nodes. The approach views control as a planning task, where nodes individually develop local plans and asynchronously exchange plan information in order to form and follow partial global plans that specify cooperative actions and interactions. In this talk, I will describe how partial global planning has been implemented in a simulated distributed problem solving network for vehicle monitoring. I will discuss experimental results showing that partial global planning improves coordination without introducing excessive overhead, allows effective coordination even in dynamically changing situations, and provides flexibility so that nodes can cooperate in different ways to achieve a variety of goals.