Prediction With Approximated Gaussian Process Dynamical Models

The modeling and simulation of dynamical systems is a necessary step for many control approaches. Using classical, parameter-based techniques modern systems, e.g., soft robotics or human–robot interaction, are often challenging even infeasible due to the complexity system dynamics. In contrast, data-driven approaches need only minimum prior knowledge scale with system. particular, Gaussian process models (GPDMs) provide very promising results complex However, properties these GP just sparsely researched, which leads “blackbox” treatment in scenarios. addition, sampling GPDMs prediction purpose respecting their nonparametric nature non-Markovian dynamics making theoretical analysis challenging. this article, we present approximated GPDMs, Markov analyze properties. Among others, error analyzed conditions boundedness trajectories provided. outcomes illustrated numerical examples that show power while computational time significantly reduced.