Skip to Main content Skip to Navigation

Localisation d'un robot mobile autonome en environnements naturels

Abstract : This thesis focuses on the problem of localizing an autonomous mobile robot in natural environments. The first part of the manuscript presents the problem in a functional aspect and explores algorithmic methods that produce an estimate of the rover position. A classification of such methods is suggested and three main classes are proposed. The first class is called "local". Algorithms of this class work directly on raw data perceived by the robot and usually run at high frequency; the rover position is computed incrementally, by summing elementary displacements. It is for instance the case of odometry, the first method developed in the manuscript and to which a chapter is devoted. An original visual motion estimation method is then proposed: it uses stereo-vision and pixel tracking in video-image sequences. This method helps to palliate odometry drawbacks, notably on uneven terrains. However, positions computed by local methods enventually drift. It is thus necessary, for long term navigation, to use methods of the second class which we call "global": such methods (such as landmark based approaches) reduce the drifts of the local methods. A method based on local digital elevation maps --- incrementally built --- is proposed in the third chapter. Digital elevation maps allow to refine the position estimate by minimizing a distance between a local 3D image and the environment's model. Furthermore, thanks to an original map structure, the rover's trajectory is memorized: this allow to back-propagate modifications on previous position estimates (corrected by landmark based algorithms for instance) and thus guarantee a better spatial coherence of the global model. The last class of algorithms is called "absolute" and concerns localization methods that work on high level data, issued of a fusion and an interpretation of local data. However, such methods are not presented in this document. The next chapter of the manuscript presents localization from the "integr ation" point of view and analyzes the problems raised by integrating several localization algorithms together. To obtain real autonomy, the rover must be able to use a large panel of functionalities (even redundant ones). Integrating these methods has been done in the LAAS architecture context (an architecture for autonomous systems). This architecture gives a mean to use modular functionalities and allows them to cooperate. It has been necessary to precise how localization functionalities had to be integrated. In particular, time-stamps, data tagging and data fusion problems have been addressed.
Document type :
Complete list of metadatas
Contributor : Emilie Marchand <>
Submitted on : Monday, February 19, 2007 - 1:24:16 PM
Last modification on : Monday, October 19, 2020 - 11:09:47 AM
Long-term archiving on: : Wednesday, April 7, 2010 - 12:12:32 AM


  • HAL Id : tel-00131779, version 1


Anthony Mallet. Localisation d'un robot mobile autonome en environnements naturels. Automatique / Robotique. Institut National Polytechnique de Toulouse - INPT, 2001. Français. ⟨tel-00131779⟩



Record views


Files downloads