Abstract
WiFi localization, the task of determining the physical location of a mobile device from wireless signal strengths, has been shown to be an accurate method of indoor and outdoor localization and a powerful building block for location-aware applications. However, most localization techniques require a training set of signal strength readings labeled against a ground truth location map, which is prohibitive to collect and maintain as maps grow large. In this paper we propose a novel technique for solving the WiFi SLAM problem using the Gaussian Process Latent Variable Model (GPLVM) to determine the latent-space locations of unlabeled signal strength data. We show how GPLVM, in combination with an appropriate motion dynamics model, can be used to reconstruct a topological connectivity graph from a signal strength sequence which, in combination with the learned Gaussian Process signal strength model, can be used to perform efficient localization.
Introduction
Indoor localization is a critical component for modern mobile computing, yet it remains challenging due to the complex nature of signal propagation in indoor environments. Traditional methods, such as WiFi fingerprinting, rely on labor-intensive manual data collection to create a "radio map." This paper addresses the WiFi Simultaneous Localization and Mapping (WiFi-SLAM) problem: building a radio map and localizing a device using only raw, unlabeled signal strength data. By treating SLAM as a manifold learning problem, we demonstrate that the latent structure of wireless signals can be leveraged to reconstruct physical geometry.
Methodology
The proposed approach leverages the Gaussian Process Latent Variable Model (GP-LVM) to treat physical coordinates as latent variables that explain observed signal strengths.
Manifold Learning with GP-LVM
The high-dimensional signal strength data is reduced to a two-dimensional latent space. Gaussian Processes (GPs) are employed to model the signal strength of each access point, providing a probabilistic representation of the environment. This allows the system to "unfold" the high-dimensional signal data into a 2D plane that preserves the relationships between data points, as shown in the manifold visualization below.
Motion Dynamics and Back-Constraints
To constrain the optimization, a motion dynamics model is integrated as a prior. This model enforces physical constraints such as velocity limits, heading consistency, and path alignment. Furthermore, back-constraints are used to ensure that similar signal observations map to nearby latent locations, which is essential for loop closure and topological consistency.
Experimental Evaluation
The system was evaluated using data collected from a university building floor. Traces ranging from 250m to 500m were recorded using an iPAQ handheld device equipped with a standard WiFi card. Ground truth paths were manually annotated for performance evaluation purposes only.
Results
The GP-LVM successfully reconstructed the topological layout of the office environment, identifying key intersections and aligned hallways. The system achieved a mean localization error of 3.97 ± 0.59 meters, which is comparable to some supervised methods despite requiring no labeled training data.
As illustrated in the comparison of signal strength surfaces, the learned GP sensor models closely approximate the models built using ground-truth data. This visual confirmation validates the latent space's ability to accurately capture the underlying physics of signal propagation.
Discussion and Limitations
While the unsupervised nature of WiFi-SLAM significantly improves scalability, the current motion model's preference for rectilinear paths may limit its effectiveness in non-standard architectural layouts. Additionally, the method's performance is tied to the density of available WiFi access points. Future research could incorporate inertial sensors, such as accelerometers or compasses, to enhance the motion model and further reduce localization error. Computational complexity also remains a factor for larger environments, suggesting the need for sparse GP techniques.
Conclusion
WiFi-SLAM demonstrates that the latent structure of WiFi signals can be used to reconstruct physical geometry without manual calibration. By shifting the view of WiFi localization from coordinate-matching to a topological problem, this research provides a scalable foundation for indoor navigation systems in complex, unmapped environments.
Summary updated