Characterizing and understanding three-dimensional natural phenomena (e.g. the energy budget of a forest canopy) requires massive parallel sensing and data processing. Static wireless sensor networks enable sampling of the environment at a finer granularity than ever before to increase our knowledge of biological phenomena. However, static networks are spatially constrained to a limited number of sampling points. A combination of actuated and static sensor networks, on the other hand, can reveal much greater detail about the phenomena. This is because a mobile node can achieve a high degree of spatial sampling and the static nodes can achieve a high degree of temporal sampling.
This dissertation presents Bioscope: a set of algorithms and techniques that enable a scientist to use an actuated wireless sensor network to systematically study biological phenomena. It consists of two major components: 1) A space-filling component, an exploration method that samples a phenomenon such that the samples spread in space with maximum inter-point distances. The primary goal of the space-filling component is to achieve a high degree of robustness for understanding the fundamentals of the phenomenon. 2) An adaptive component, which generates a model of the phenomenon through runtime adaptation and orchestrates sample collection such that the performance of the estimated model is maximized. The combination of these two approaches enables the scientist to efficiently observe the underlying phenomenon.
This thesis presents statistical techniques that form the skeleton of a data collection experiment using an actuated sensor network and discusses the necessary modifications and design choices to adapt such schemes to the constraints of our problem. It describes our approach toward the problem and provides experimental evidences that demonstrate the applicability of our approach.