Set-Membership Approach to Map-Based Localization

Benoît Desrochers (1)

1 LAAS-RIS - Équipe Robotique et InteractionS

Absolute position estimation is a key function of autonomous systems and is often required at the mission level. One way the get such information is to use an initial model of the environment provided for instance, by a geographic database or built by an other robot. This study aims at providing an algorithm which matches the output of a sensor with an initial model to estimate the pose of the robot in a guaranteed way, using the intervals analysis framework. By construction, initial models do not perfectly fit the reality and the acquire data set can contains an unknown, and potentially large, percentage of outliers. When the environment is described by a surface, the set membership estimator \textit{GOMNE} can be used to concurrently estimate the number of outliers and the localisation parameters. However, in the general case, with full 3D representation and partially mapped objects, it can not be used. To cope this issue, a new algorithm called Outer-GOMNE is proposed. By combining intervals methods with local estimation algorithms, it have been applied to the localization problem. After a simulated test case with a 2D map, an experimental validation using real laser data and different 3D models is reported to illustrate the performance of the method. Results are compared with ground truth provided by a differential GPS. Outer-GOMNE is able to robustly enclose the ground truth in a sub-paving (union of non overlapping boxes)

Mots clés: Robotics; Robot localization; set-membership estimation; interval analysis

Domaines: Robotique [cs.RO]

Benoît Desrochers. Set-Membership Approach to Map-Based Localization. 2014. ⟨hal-01065126⟩