Probabilistic modeling of planar pushing
Name
1057285378-MIT.pdf
Description
Full printable version
Size
5.37 MB
Format
Adobe PDF
Checksum (MD5)
4de6a60341d24222fab9ffc25be2c9a5
Author(s)
Bauza Villalonga, Maria
Advisor(s)
Alberto Rodriguez.
Date Issued
2018
Publisher
Massachusetts Institute of Technology
Abstract
This work studies the problem of data-driven modeling and stochastic filtering of complex dynamical systems. The main contributions are GP-SUM, a filtering algorithm tailored to systems expressed as Gaussian processes (GP), and the probabilistic modeling of planar pushing by combining input-dependent GPs and GP-SUM. The main advantages of GP-SUM for filtering are that it does not rely on linearizations or unimodal Gaussian approximations of the belief. Moreover, it can be seen as a combination of a sampling-based filter and a probabilistic Bayes filter as GP-SUM operates by sampling the state distribution and propagating each sample through the dynamic system and observation models. Effective sampling and accurate probabilistic propagation are possible by relying on the GP form of the system, and a Gaussian mixture form of the belief. In this thesis we show that GP-SUM outperforms several GP-Bayes and Particle Filters on a standard benchmark. To characterize the dynamics of pushing, we use input-dependent GPs to learn the motion of the pushed object after a short time step. With this approach we show that we can learn accurate data-driven models that outperform analytical models after less than 100 samples and saturate in performance with less than 1000 samples. We validate the results against a collected dataset of repeated trajectories, and use the learned models to study questions such as the nature of the variability in pushing, and the validity of the quasi-static assumption. Finally, we illustrate how our learned model for pushing can be combined with GP-SUM, and demonstrate that we can predict heteroscedasticity, i.e., different amounts of uncertainty, and multi-modality when naturally occurring in pushing.
Description
Thesis: S.M., Massachusetts Institute of Technology, Department of Mechanical Engineering, 2018.
Cataloged from PDF version of thesis.
Includes bibliographical references (pages 65-68).
Subjects
Mechanical Engineering.
MIT Department
Massachusetts Institute of Technology. Department of Mechanical Engineering
Terms of Use
MIT theses are protected by copyright. They may be viewed, downloaded, or printed from this source but further reproduction or distribution in any format is prohibited without written permission.
Persistent DSpace Link
