Skip to Main content Skip to Navigation
Theses

Deep learning, inertial measurements units, and odometry : some modern prototyping techniques for navigation based on multi-sensor fusion

Abstract : This thesis deals with state estimation for vehicles that are equipped with various sensors such as cameras and inertial measurement units. The Kalman filter is a widely used tool that estimates the state of a dynamical system, which raises theoritical questions for the nonlinear systems present in navigation, and that relies on physical models and parameters that need to be efficient tuned by the user. This thesis contributes to Kalman filtering for navigation, where the contributions focus on two different manners. The first manner consists in building on the recent invariant extended Kalman filter to address challenging issues, namely the inconsistency of extended Kalman filter for the problem of simultaneous localization and mapping, navigation with vision sensors, and the more general question of Kalman filtering on manifolds. The second manner consists in using recent tools from the field of artificial intelligence, namely deep learining, to improve Kalman filters, notably to relate sensors’ measurements to the state, and to tune the filter efficiently, that is, using machine learning techniques to find a dynamical tuning strategy of the Kalman filter parameters that best matches the data and that is also able to provide new information to the filter. The thesis thus introduces differents filtering algorithms and deep neural networks whose implementation are made open-source.
Document type :
Theses
Complete list of metadata

https://pastel.archives-ouvertes.fr/tel-03262132
Contributor : Abes Star :  Contact Connect in order to contact the contributor
Submitted on : Wednesday, June 16, 2021 - 12:37:08 PM
Last modification on : Tuesday, July 13, 2021 - 3:09:11 AM
Long-term archiving on: : Friday, September 17, 2021 - 6:42:07 PM

File

2020UPSLM067_archivage.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03262132, version 1

Citation

Martin Brossard. Deep learning, inertial measurements units, and odometry : some modern prototyping techniques for navigation based on multi-sensor fusion. Machine Learning [cs.LG]. Université Paris sciences et lettres, 2020. English. ⟨NNT : 2020UPSLM067⟩. ⟨tel-03262132⟩

Share

Metrics

Record views

147

Files downloads

162