Precise and accurate wireless signal strength mappings using Gaussian processes and path loss models

Renato Miyagusuku, Atsushi Yamashita, Hajime Asama

Producción científica: Contribución a una revistaArtículorevisión exhaustiva

17 Citas (Scopus)

Resumen

In this work, we present a new modeling approach that generates precise (low variance) and accurate (low mean error) wireless signal strength mappings. In robot localization, these mappings are used to compute the likelihood of locations conditioned to new sensor measurements. Therefore, both mean and variance predictions are required. Gaussian processes have been successfully used for learning highly accurate mappings. However, they generalize poorly at locations far from their training inputs, making those predictions have high variance (low precision). In this work, we address this issue by incorporating path loss models, which are parametric functions that although lacking in accuracy, generalize well. Path loss models are used together with Gaussian processes to compute mean predictions and most importantly, to bound Gaussian processes’ predicted variances. Through extensive testing done with our open source framework, we demonstrate the ability of our approach to generating precise and accurate mappings, and the increased localization accuracy of Monte Carlo localization algorithms when using them; with all our datasets and software been made readily available online for the community.

Idioma originalInglés
Páginas (desde-hasta)134-150
Número de páginas17
PublicaciónRobotics and Autonomous Systems
Volumen103
DOI
EstadoPublicada - may. 2018
Publicado de forma externa

Huella

Profundice en los temas de investigación de 'Precise and accurate wireless signal strength mappings using Gaussian processes and path loss models'. En conjunto forman una huella única.

Citar esto