UWB Indoor Positioning: A Comparative Study of Machine Learning Models for Precise Robot Localization

Abstract

Indoor positioning has long been a challenging problem, with traditional Global Navigation Satellite Systems (GNSS) struggling indoors due to signal attenuation, multipath effects, and the inability of satellite signals to penetrate dense building materials such as concrete and metal. Ultra-Wideband (UWB) technology has emerged as a game-changer, offering high accuracy, low power consumption, and robustness to obstacles, making it ideal for indoor positioning applications. This study explores the integration of machine learning (ML) techniques with UWB technology to enhance indoor positioning accuracy by testing the performance of six Machine Learning regression models for predicting the 3D coordinates (estX, estY, estZ) of a robot in a challenging indoor environment. The results demonstrate that ensemble methods, particularly Random Forest and Gradient Boosting, outperform other models, achieving the lowest error, while simpler models like Linear Regression struggle to handle the complexities of indoor environments. The Decision Tree model offers a balance between simplicity and accuracy, while SVR and ANN deliver competitive results, with ANN showing promise in handling high-dimensional data. Notably, this direct ML approach not only matches but often surpasses traditional methods such as Kalman filters and kinematic models in terms of accuracy, and complexity. These findings highlight the transformative potential of combining ML with UWB technology, offering a robust and scalable solution for indoor positioning with promising applications in smart buildings, asset tracking, and indoor navigation, paving the way for more reliable and efficient systems in Industry 4.0 and IoT. Each of the tested models brings different strengths and potential challenges to the system, and their evaluation in the context of UWB data for robot localization promises insightful findings into the most effective approaches for this application.