AE Seminar

Leveraging and Sensing of Dynamical Structure in Data
Assimilation for Nonlinear Systems

ft

Ryne Beeson

Assistant Professor | Mechanical and Aerospace Engineering | Princeton University.

Thursday, April 3
2 - 3 p.m.
Klaus Advanced Computing 2443
 

About the Seminar: 
The need to perform data assimilation is ubiquitous in modern life, as evidenced by the proliferation of sensors on systems both small (e.g., UAVs and CubeSats) and within vast ones (e.g., atmosphere, oceans, and near-Earth space). Data assimilation requires specification of a dynamical model, which may be stochastic, nonlinear, and have varying spatial-temporal scales. The variety of problems means that there is no single best solution for them all, but whenever there is known or detectable structure for a given problem, we would be wise to make use of it. In this talk, we consider two scenarios: the first in which we can generate approximations of important dynamical structures that separate the flow, and the second where we cannot explicitly construct an approximation, but we can use an optimal control approach to indirectly sense similar structure. The application we consider for the first scenario is the problem of space situational awareness in the cislunar domain. Here we propose to make greater use of the explicit knowledge of dynamical structures in local neighborhoods of special solutions (equilibria, periodic orbits, etc.) to construct a new parameterized distribution that is naturally partitioned to the dynamical flow. In the second scenario we investigate the efficacy of using an optimal control approach to particle filtering, where particles are guided toward the observation neighborhood while maintaining approximate model consistency. To overcome issues of particle degeneracy that arise, we propose the use of an intermediate resampling approach. We will conclude the talk with some thoughts on how the aforementioned approaches can be further evolved to enable more efficient, stable, and high performing data assimilation for highly nonlinear and chaotic systems.
 

About the Speaker: 
Ryne Beeson is an Assistant Professor of Mechanical and Aerospace Engineering at Princeton University. He received a Ph.D. in Aerospace Engineering from the University of Illinois at Urbana-Champaign (2020) and was awarded the Kenneth Lee Herrick Memorial Award for his dissertation research. He also holds an M.S. in Mathematics and B.S, M.S. in Aerospace Engineering from Illinois. Prior to starting at Princeton University in 2021, he was a Senior Scientist for CU Aerospace (CUA) L.L.C. (2016-2021) in Champaign, Illinois. He was PI for several NASA Phase I and II SBIRs that developed astrodynamics related software, including the multibody trajectory optimization software (pydylan), which continues to be advanced with his group at Princeton. He is a recent awardee of the Air Force Office of Scientific Research Young Investigator Program and is a member of the AIAA Guidance Navigation & Control Committee and American Astronautical Society Technical Committee.