Wöber, Wilfried
Novotny, Georg
Aburaia, Mohamed
Otrebski, Richard
Kubinger, Wilfried
2018-08-01
<p>Localization in mobile robotics is an active research area. Statistical tools such as Bayes filters are used for localization. The mplementation of Gaussian processes in Bayes filters to estimate transition and measurement models were introduced recently. The non-linear and non-parametric nature of Gaussian processes leads to new possibilities in modelling systems. The high model complexity and computation expense based on the size of the dataset are shortcomings of Gaussian process Bayes filters. This work discusses our approach of a sparsing process of a dataset based on Bayesian information criterion model selection and global optimization. The developed approach combines the idea of avoiding model overfitting and Bayesian optimization to estimate a sparse representation of a Gaussian process. Based on visual odometry data of a mobile robot, the method was evaluated. The results show the operability of the system and unfold limitations of the current implementation such as random-initialization. </p>
https://doi.org/10.5281/zenodo.3402876
oai:zenodo.org:3402876
eng
Zenodo
https://zenodo.org/communities/fhtw
https://doi.org/10.5281/zenodo.3402875
info:eu-repo/semantics/openAccess
Creative Commons Attribution 4.0 International
https://creativecommons.org/licenses/by/4.0/legalcode
Austrian Robotics Workshop 2018
Localization
Mobile Robotics
Modelling
Sparsing
Estimating a Sparse Representation of Gaussian Processes Using Global Optimization and the Bayesian Information Criterion
info:eu-repo/semantics/conferencePaper