Automatic construction, maintenance, and optimization of dynamic agent organizations.

摘要

The goal of this dissertation is to generate organizational structures that increase the overall performance of a multiagent coalition, subject to the system's complex coordination requirements and maintenance of a certain operating point. To this end, a generalized framework capable of producing distributed approximation algorithms based on the new concept of multidirectional graph search is proposed and applied to a family of connectivity problems. It is shown that a wide variety of seemingly unrelated multiagent organization problems live within this family. Sufficient conditions are identified in which the approach is guaranteed to discover a solution that is within a constant factor of the cost of the optimal solution. The procedure is guaranteed to require no more than linear---and in some well defined cases logarithmic---communication rounds. A number of examples are given as to how the framework can be applied to create, maintain, and optimize multiagent organizations in the context of real world problems. Finally, algorithmic extensions are introduced that allow for the framework to handle problems in which the agent topology and/or coordination constraints are dynamic, without significant consequences to the general runtime, memory, and quality guarantees.