Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 17: Modeling and Data Analysis

DY 17.5: Talk

Tuesday, March 17, 2015, 10:30–10:45, BH-N 333

Learning equations of motion from sparse observations — •Andreas Ruttor, Philipp Batz, and Manfred Opper — Technische Universität Berlin

Equations of motion describe the dynamics of a system in terms of differential equations. These can be derived from theory if all the relevant properties are exactly known. But for real devices, e.g. robots, this is usually not the case. Instead one can drive the system applying a noisy control force and learn the equations of motion by observing its behavior. For that purpose we use a non-parametric approach based on Gaussian process regression, which does not require a detailed model of the dynamics, but still allows to include prior knowledge. As our method is based on estimating the probability distribution in phase space, it works with sparse observations, where the time intervals between data points are large.