The ultimate goal of this research work is to optimally schedule measurement acquisition process in order to maximize the collective information of the spatio-temporal phenomenon of interest. The particular applications of interest is the tracking of toxic material plumes subsequent to natural or man-made disasters (e.g. Eyjafjallajo ̈kull eruption, Deepwater Horizon oil-spill) and geomagnetic survey for GEOINT application by using optimally designed, static or dynamic sensing architectures. Autonomous agents such as Unmanned Air Vehicles (UAVs) equipped with various sensors are complex sensing systems and can gather big data (enormous amount of data collected from onboard sensors such as Radar, visible and infrared sensors1) for surveillance, mapping and management of risk in various applications. Due to predominant local behavior of most sensors (e.g. all cameras have with limited field of view), the dynamic sensing problem is of particular interest to make the best possible measurement, as “better” measurements can lead to better characterization of the underlying process.
To this end, the optimal sensing problem is posed as a stochastic optimal control problem to find optimal location of sensors and sensor modalities to tradeoff between competing indices such as fuel consumption or coverage of domain and the information gathered. An effective sensor planning is contingent upon meaningful information measures that govern the choice of sensor observations to make meaningful model forecasts. Information theoretic sensor planning can be viewed as a statistically informed investing process that aims at maximizing the prospective information gain by the collective observations obtained from the sensor network.