MAE Seminar: New Methods for State Estimation and Adaptive Observation of Environmental Flow Systems Leveraging Coordinated Swarms of Sensor Vehicles
University of California, San Diego
Flow Control & Coordinated Robotics Labs
Abstract: Accurate long-term forecasts of the path and intensity of hurricanes are imperative to protect property and save lives. Accurate estimations and forecasts of the spread of large-scale contaminant plumes, such as those from Deepwater Horizon, Fukushima and recent volcanic eruptions in Iceland, are essential for assessing environment impact, coordinating remediation efforts and in certain cases moving folks out of harm’s way. The challenges in estimating and forecasting such systems include: (a) environmental flow modeling, (b) high-performance real-time computing, (c) assimilating measured data into numerical simulations and (d) acquiring in-situ data, beyond what can be measured from satellites, that is maximally relevant for reducing forecast uncertainty. This talk will focus on new techniques for addressing (c) and (d), namely, data assimilation and adaptive observation, in both hurricanes and large-scale environmental plumes. In particular, we will present a new technique for the energy-efficient coordination of swarms of sensor-laden balloons for persistent, in-situ, distributed, real-time measurement of developing hurricanes, leveraging buoyancy control only (coupled with the predictable and strongly stratified flowfield within the hurricane). Animations of these results are available at http://flowcontrol.ucsd.edu/3dhurricane.mp4 and http://flowcontrol.ucsd.edu/katrina.mp4. We also will survey our unique hybridization of the venerable Ensemble Kalman and Variational approaches to large-scale data assimilation in environmental flow systems, and how essentially the dual of this hybrid approach may be used to solve the adaptive observation problem (that is, the optimal routing of sensor vehicles) in a uniquely effective and rigorous fashion.
Biography: Tom Bewley (B.S./M.S. Caltech 1989, diploma VKI 1990, Ph.D. Stanford 1998) directs the Flow Control and Coordinated Robotics Labs at UCSD. His flow control group, established in 1998, develops new algorithms for high-performance computing, optimization and control of PDE systems, data assimilation and adaptive observation of environmental flows. His robotics group, established in 2005, couples minimalist mechanical design with robust low-cost manufacturing and modern cellphone electronics & software to develop novel robotic systems in the surveillance, educational and consumer sectors, and to deliver them to market. These two groups work synergistically on projects involving the optimal routing of robotic sensor vehicles for real-time estimation and forecasting of environmental flows, as he will discuss in his talk.