This project investigates distributed algorithms for organizing, structuring, and deploying sensor networks so that they can accomplish a variety of high-level global estimation tasks. Special attention is given to probabilistic methods that can deal with sensor noise, signal interference, and measurement inaccuracies. The overall goal is to encapsulate a small number of paradigms for sensor collaboration, information aggregation, and distributed reasoning that can serve a wide variety of applications and ease the burden of implementing this type of networked software anew each time. Communication patterns are explored that combine sensing, information exchange, and processing actions that jointly perform the required estimation in a lightweight, robust, and uncertainty-aware manner. The net effect is to augment the network stack above the data link layer with another layer that is data-driven and performs collaborative reasoning, exploiting sensed information and its value to the network task. Such task-driven communication patterns, once they have been identified and implemented, will facilitate the quick deployment of sensor networks for new tasks and seed additional applications. The collaboration and reasoning problems addressed are important as well for other distributed systems, including distributed data-bases and peer-to-peer networks.
Research on this project will combine design, simulation, theoretical analysis, and the implementation of an experimental sensor network testbed within a university building for exploring and validating these ideas.
