Adaptive multiagent planning in a distributed environment

Abstract

Currently there are several paradigms for distributed problem solving. First, the Contract Nets [ 19, 51 approach emphasizes the task-sharing and result-sharing phases of distributed problem solving through formal contracting procedures, including task announcement, bidding, and contract award. Second, the Actors style of concurrent programming [ 15, 1] provides the ability to model shared objects which have changing local states, to reconfigure dynamically the network, and to promote inherent parallelism in an application. Third, the functionally accurate, cooperative (FA/C) distributed system [16, 7, 81 is characterized by its handling of distribution-caused uncertainty and errors. In a FA/C system partial tentative results are exchanged among nodes, within the context of common goals. Fourth, when applied to multiagent planning, nodes channel their results to coordinating nodes that generate and distribute multiagent plans; see [ 11, 12, 131. We have constructed a new model for distributed problem solving based on parallel logic programming; see IZ 3, 6 71.