Guaranteed Simultaneous Localization And Mapping (SLAM) for Vehicles

Retour à la liste des thèses
Ajouté le: 26/05/2014
Directeur : LAMBERT Alain -
Titre : Guaranteed Simultaneous Localization And Mapping (SLAM) for Vehicles
Thèmes : Automatique, Signal, Télécoms, Systèmes embarqués
Laboratoires : IEF - Institut d'Electronique Fondamentale UMR 8622
Description :

In order to move safely, a vehicle must have an embedded system (composed of an electronic and software architecture) able to provide relevant information on its positioning and the environment. This thesis aims to design a bounded-error embedded system that will provide a guaranteed Simultaneous Localization And Mapping.


The localization of a vehicle consists in determining its position from the measured distances to known landmarks (feature points in the image and / or satellite data). Simultaneous Localization And Mapping (SLAM) algorithms aim to build an environment map and to estimate the robot pose in the same time. Many researches were conducted to develop SLAM algorithms like EKF-SLAM (Extended Kalman Filter for Simultaneous Localization And Mapping), FAST SLAM, GRAPH SLAM, DP-SLAM which aim to improve consistency, accuracy or robustness. Other algorithms derivate from the EKF-SLAM, such as algorithms using Unscented Kalman Filter (UKF) which increase the localization accuracy against the classical EKF algorithm based on a linearized model.

Although Bayesian localization and mapping algorithms were widely used and improved, they have some drawbacks due to non linearity of their models and Gaussian hypothesis of noises. For instance, Bayesian approaches suffer from consistency problems (lack of realism on the positioning uncertainties) that are not resolved to this day.

We are developing an alternative approach based on Bounded-Error State Estimation that does not need linearization. Sensor data are assumed to belong to an interval without any assumption on the noise distribution within this interval. Guaranteed localization based on interval analysis is a key to safer autonomous driving.
We obtained preliminary results in an indoor environment for a mobile robot embedding a camera and wheels encoders. This study will be extended to other sensors (IMU and GPS) and a more complex vehicle dynamic.
The main effort of this thesis will be to develop a new Bounded-Error SLAM algorithm and optimize its computational time.


PhD supervisor (contact person): Alain Lambert