This project aims to develop a systematic framework and scalable computational techniques for sensor placement and time-critical sensor motion strategies for dynamically evolving, spatially-distributed field quantities. Such fields describe environmental variables such as plume concentrations, pressures, wind velocity, or ocean salinity and temperature as is common in techniques of data assimilation. The increasing ubiquity of mobile sensing platforms presents an opportunity to improve upon traditional estimation and prediction techniques in data assimilation. A major goal of this project is to discover how a limited number of sensors should be placed or maneuvered to optimize estimation and prediction fidelity in a time-critical manner. The results of this project will enable new design techniques that can ultimately aid in natural disaster prediction and response management. From forest fire fronts, to floods and other severe weather events, data assimilation techniques are currently indispensable for prediction. They are however typically constrained by the use of only the momentarily available environmental measurements. This research would produce a systematic methodology for proactive, optimal dispatching and trajectory planning of mobile sensors whose measurements can then reduce prediction uncertainty for critical regions and quantities. This would be a significant aid to natural disaster response preparation and management. The success of this project will not only promote the fundamental science in estimation and predictive technologies but also help advance the nation?s disaster prediction and response capability.
A model-based estimation and prediction approach is adopted, where the use of underlying physical laws enable high resolution estimation and prediction of spatio-temporally varying physical fields from sparse and limited measurements. The main thrust of the project is the development of a new framework of dynamic exploration, in which sensors' motion and/or location is designed using optimal and feedback control techniques with the objective of maximizing information gain metrics. While heuristics can be easily developed in individual settings, there is a need for a systematic theory of motion control design for the purpose of estimator optimization, especially in dynamic environments. Thus the sensor motion problem in an unknown environment is reformulated as an optimal control problem with the objective being the maximization of information reward metrics, or minimization of error covariances. New techniques will need to be developed to address these non-traditional optimal and feedback control problems. Large-scale computational issues such as low-rank approximations of error covariances will be explored and utilized. The project will address specific research questions that arise due to differences between sensing modalities such as point-wise versus tomographic or aggregates sensing.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
