Fusion multiphysique pour système de localisation indoor

This work is part of a Cifre thesis carried out in collaboration between the SATIE laboratory and the ez-Wheel company. The company specializes in developing compact, integrated motorized wheels that facilitates the motorization and the automation of mobile platforms aimed at transporting heavy loads. Throughout the course of this project, the company introduced a new generation of products tailored for safe mobile robotics applications in industrial environments. A critical requirement for developing an autonomous mobile robot based on these wheels is a localization system that is suitable for the targeted industrial environments. To address this requirement, we initially investigated the localization approaches proposed in the scientific literature. Our first proposed solution is based on multi-LiDARs fusion while assuming a static or weakly dynamic environment. However, the targeted industrial environments are expected to have highly dynamic surroundings. Therefore, after observing that in such environments, the ceiling remains invariant and mostly static, we introduced a new approach that separates the navigation space from the localization space. Thus, to detect obstacles and ensure safe movements, we utilize a 2D LiDAR on the horizontal plane. While we use a camera oriented towards the ceiling (vertical plane) to guarantee a robust relative localization. This decoupling effectively eliminates the issues related to the detection and the filtering of dynamic objects, thereby improving the localization quality without extra processing cost. We have proposed a ceiling-vision localization system based on the Direct Sparse Odometry (DSO), and validated its performance and accuracy on our experimental platform. Finally, to further validate our approach in real-world scenarios, we designed and conducted an experiment to collect a multisensor dataset focused on ceiling-vision. The novelty of this work resides in the fact that there are no available datasets in the scientific literature that allow the evaluation of ceiling-camera localization methods. The methodological and technological choices of this work were strongly influenced by the industrial context of the thesis, aiming to present a TRL 6 prototype for subsequent integration into the company's products.