eduzhai > Applied Sciences > Engineering >

Scalable Identification of Partially Observed Systems with Certainty-Equivalent EM

  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: System identification is a key step for model-based control, estimatordesign, and output prediction. This work considers the offline identificationof partially observed nonlinear systems. We empirically show that thecertainty-equivalent approximation to expectation-maximization can be areliable and scalable approach for high-dimensional deterministic systems,which are common in robotics. We formulate certainty-equivalentexpectation-maximization as block coordinate-ascent, and provide an efficientimplementation. The algorithm is tested on a simulated system of coupled Lorenzattractors, demonstrating its ability to identify high-dimensional systems thatcan be intractable for particle-based approaches. Our approach is also used toidentify the dynamics of an aerobatic helicopter. By augmenting the state withunobserved fluid states, a model is learned that predicts the acceleration ofthe helicopter better than state-of-the-art approaches. The codebase for thiswork is available at this https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×