Seminar: Research Themes at ARC for Autonomy in Data-Driven and Complex Systems

In this talk, we will look at a snapshot of the new Laboratory for Autonomy in Data Driven and Complex Systems (LADDCS) at ARC. LADDCS focuses on theoretical and computational research in uncertainty characterization, forecasting and fusion, as well as control of uncertain (better known as stochastic) systems. Challenges associated with the Fokker-Planck equation (FPE), a holy grail problem in stochastic dynamics will be discussed. Two fundamentally distinct solution paradigms will be presented – i.) tensor-based discretization, and, ii.) adaptive particle discretization in the framework of Monte Carlo methods. The key objective of both classes of methods is to accurately capture non-Gaussianity while alleviating the curse of dimensionality. We will present our results in a wide variety of fields, including weather-forecasting, space-situational awareness and the modeling of polymeric fluids.

About the Speaker

Time permitting, we will talk about another topic of interest at LADDCS, namely Chance-Constrained Optimal Control (CCOC) and some representative problems in this domain. We will discuss why CCOC is an important alternative to Robust Control and the challenges associated with its solution. Some initial results with a new semi-analytic parametric surrogate called the split-Bernstein approximation will be presented.