Attentional Graph Meta-Learning for Indoor Localization Using Extremely Sparse Fingerprints

TL;DR

This paper tackles fingerprint-based indoor localization under extremely sparse fingerprints by marrying an Attentional Graph Neural Network (AGNN) with Model-Agnostic Meta-Learning (MAML) in a framework called Attentional Graph Meta-Learning (AGML). It introduces a Distance Learning Module (DLM) and an Adjacency Learning Module (ALM) to infer a meaningful graph structure from limited CSI/CIR data, and adds two data-augmentation strategies: unlabeled fingerprints from moving platforms and synthetic labeled fingerprints via digital twins with distribution alignment. AGML is trained on synthetic environments to learn robust meta-parameters and then adapted to real environments, enabling rapid localization of unlabeled TPs with minimal labeled data. Experimental results on synthetic and real datasets (including consumer-grade WiFi and professional equipment) show that AGML consistently outperforms baselines, achieving lower RMSE and faster convergence, thereby reducing labor costs for fingerprint collection and improving scalability in diverse indoor environments.

Abstract

Fingerprint-based indoor localization is often labor-intensive due to the need for dense grids and repeated measurements across time and space. Maintaining high localization accuracy with extremely sparse fingerprints remains a persistent challenge. Existing benchmark methods primarily rely on the measured fingerprints, while neglecting valuable spatial and environmental characteristics. In this paper, we propose a systematic integration of an Attentional Graph Neural Network (AGNN) model, capable of learning spatial adjacency relationships and aggregating information from neighboring fingerprints, and a meta-learning framework that utilizes datasets with similar environmental characteristics to enhance model training. To minimize the labor required for fingerprint collection, we introduce two novel data augmentation strategies: 1) unlabeled fingerprint augmentation using moving platforms, which enables the semi-supervised AGNN model to incorporate information from unlabeled fingerprints, and 2) synthetic labeled fingerprint augmentation through environmental digital twins, which enhances the meta-learning framework through a practical distribution alignment, which can minimize the feature discrepancy between synthetic and real-world fingerprints effectively. By integrating these novel modules, we propose the Attentional Graph Meta-Learning (AGML) model. This novel model combines the strengths of the AGNN model and the meta-learning framework to address the challenges posed by extremely sparse fingerprints. To validate our approach, we collected multiple datasets from both consumer-grade WiFi devices and professional equipment across diverse environments. Extensive experiments conducted on both synthetic and real-world datasets demonstrate that the AGML model-based localization method consistently outperforms all baseline methods using sparse fingerprints across all evaluated metrics.

Figures (18)

• Figure 1: Two fingerprint augmentation strategies based on the original real-world indoor localization scenario.
• Figure 2: Model architecture of the tailored AGNN for indoor localization, which is composed of a DLM, an ALM, and MGALs.
• Figure 3: The framework of the AGML model, which consists of a meta-training phase on synthetic datasets and a meta-test phase on the real-world dataset. In meta-training, three steps are performed: task definition ($m$ synthetic datasets are randomly sampled into $m\times r$ tasks), inner loop update (the model performs one-step gradient descent on each task's support set), and outer loop update (meta-parameters are optimized across multiple tasks' query sets). In meta-testing, the model undergoes: distribution alignment (aligning the feature distributions of the real-world dataset with synthetic datasets), model adaptation (updating the meta-parameters using the transformed dataset), and localization of TPs (locating all unlabeled TPs).
• Figure 4: Photographs of the Hall-scenario and Lab-scenario on the CUHKSZ campus.
• Figure 5: Simulations of the Hall-scenario and Lab-scenario on the CUHKSZ campus, which include the same uplink communication system and AP positions as the real world. The blue area represents clutter, the orange areas represent desks, the red dots indicate receiver nodes, and the yellow dots indicate transmitter nodes. Additionally, some people are included in the simulation to better mimic real-world conditions.