Framework multissensorial robusto para localização de robôs móveis em ambientes internos

Abstract

Autonomous mobile robots have been widely employed in indoor environments, such as warehouses, hospitals, laboratories, industrial facilities, and intralogistics systems. However, reliable localization of these systems remains a challenge, especially in scenarios without Global Positioning System signal availability and subject to odometric drift, repetitive geometries, inadequate lighting conditions, and uncertain sensory measurements. Although Simultaneous Localization and Mapping (SLAM) approaches based on the Extended Kalman Filter (EKF) are frequently used for sensor fusion, their performance may be degraded by modeling errors, local linearizations, and imperfect noise characterization. In view of these limitations, this dissertation proposes a robust multisensor framework to improve the localization of mobile robots in structured indoor environments. The approach integrates wheel odometry, observations of ArUco fiducial markers, and auxiliary pose estimates obtained from Adaptive Monte Carlo Localization (AMCL). The objective is to increase the accuracy, consistency, and robustness of pose estimation under real operating conditions, in which the nominal models of the system and sensors are subject to uncertainties. To this end, a formulation based on the Robust Kalman Filter (RKF), originally developed for discrete-time linear systems with norm-bounded parametric uncertainties, is adapted to a nonlinear EKF-SLAM structure with fiducial markers. The validation is carried out through real experiments and simulations, using a mobile robot equipped with 2D and 3D LiDAR-based sensors, an RGBD camera, and ArUco markers distributed in an indoor environment without GPS signal. The results are compared with BDUE-SLAM and conventional EKF-SLAM, considering trajectory estimation, drift reduction, and localization consistency. The results indicate that the incorporation of robust estimation strategies reduces the sensitivity of the system to model and measurement uncertainties, contributing to more reliable localization in structured indoor environments. Thus, this dissertation contributes a robust localization framework for autonomous mobile robots, preserving the recursive and computationally efficient structure of Kalman filters while extending their applicability to real-world indoor navigation scenarios.