| dc.description.abstract | The construction of a collaborative sensor web that integrates sensor networks around the globe promises unprecedented advantages for today's scientists in studying and predicting weather, natural disasters, and climate change. A particularly challenging problem in the design of a global sensor web is coordination and control of these heterogeneous and distributed resources for adaptive collaboration in achieving complex science tasks. To address this problem, this dissertation presents the design and integration of novel coordination and adaptation mechanisms in a multi-agent system. This system, the Multi-agent Architecture for Coordinated Responsive Observations (MACRO), provides a powerful framework for the coordination, control, and adaptive operation of a global sensor web.
We present three major contributions to research in the fields of multi-agent systems and autonomous planning and scheduling. First, we provide a novel multi-agent coordination mechanism based on brokered task auctions with efficient subcontracting. We present the results of experiments verifying the scalability and performance of this mechanism in yielding fair and efficient allocations of complex, hierarchically-decomposable, sensor web tasks. Further, these results suggest the incorporation of an appropriate decommitment mechanism to improve task allocation performance for long-running sensor web tasks.
Next, we present the Spreading Activation Partial Order Planner (SA-POP), which is a novel decision-theoretic planning and scheduling service for local sensor network agents operating on shared computational resources. Experiments verify the ability of this service to produce high expected utility plans even under resource and scheduling constraints. Finally, we present the integration of task allocation and planning/scheduling mechanisms through agent design and organization, definition of standardized task meta-data, and agent coordination protocols. We verify the system integration through additional experiments and a case study. Experimental results illustrate the efficiency of context-sensitive planning/scheduling coordination. Further, the case study illustrates the ability of an integrated MACRO system to provide effective coordination, control, and dynamic adaptation in a global sensor web. | |
