Machine Learning Integration in Ultra-Wideband-Based Indoor Positioning Systems: A Comprehensive Review

Abstract

Ultra-Wideband (UWB) technology enables centimeter-level indoor positioning, but it remains highly sensitive to channel dynamics, multipath and Non-Line-of-Sight (NLoS) propagation. Recent studies increasingly apply Machine Learning (ML) methods to address these issues by modeling nonlinear channel behavior and mitigating ranging bias. This paper presents a comprehensive review and provides a critical synthesis of 169 research works published between 2020 and 2024, offering an integrated overview of how ML techniques are incorporated into UWB-based Indoor Positioning Systems (IPSs). The studies are grouped according to their functional objective, learning algorithm, network architecture, evaluation metrics, dataset, and experimental setting. The results indicate that most approaches apply ML to channel classification and ranging error mitigation, with Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and hybrid CNN–Long Short-Term Memory (LSTM) architectures being among the most common choices due to their ability to capture spatial and temporal patterns in the Channel Impulse Response (CIR). Despite the reported accuracy improvements, scalability and cross-environment generalization remain open challenges, largely due to the scarcity of public datasets and the lack of standardized evaluation protocols. Emerging research trends highlight growing interest in transfer learning, domain adaptation, and federated learning, along with lightweight and explainable models suitable for embedded and multi-sensor systems. Overall, this review summarizes the progress made in ML-driven UWB localization, identifies current gaps, and outlines promising directions toward more robust and generalizable indoor positioning frameworks.