- Keywords: Machine Learning, Gaussian Processes, Computational Flow Dynamics, Controls
- TL;DR: We can find good paths for mobile agents seeking to learn the state of a dynamic system using Evolving Gaussian Processes
- Abstract: We present a new method for planning trajectories of moving agents with the goal of deriving the state of a spatiotemporally evolving dynamical systems. This method is based on the spectral analysis of the linear dynamical layer of an Evolving Gaussian Processes (E-GP) model. First, we present a new algorithm for clustering elements of the system into separate invariant subspaces. Next, we discuss the relative importance of several factors in generating and choosing trajectories that help a moving Bayesian state estimator converge to an accurate state estimate quickest. Lastly, we provide preliminary results for these new methods using synthetic and real-world data sets.