Dynamic Accuracy Estimation in a Wi-Fi-based Positioning System

TL;DR

The study tackles the lack of per-result uncertainty in indoor Wi‑Fi RTLS by proposing Dynamic Accuracy Estimation (DAE), a data-driven approach that predicts the error $\delta_{\rm est}$ using the same measurements used for positioning and compares it to the true error $\delta_{\rm pos}$. Among tested regressors, random forest regression achieved the best MAE (~$0.73\,\mathrm{m}$) on robot-collected data, with a moderate correlation ($r \approx 0.5$) between $\delta_{\rm est}$ and $\delta_{\rm pos}$. The method shows improved DAE quality when incorporating the estimated smartphone location, but performance degrades in markedly different propagation conditions (e.g., body shadowing), suggesting the need for adaptation or extra features and loss penalties. This work enables real-time, uncertainty-aware localization in Wi‑Fi RTLS and provides a basis for future improvements in signal-parametric DAE and loss-function design.

Abstract

The paper presents a concept of a dynamic accuracy estimation method, in which the localization errors are derived based on the measurement results used by the positioning algorithm. The concept was verified experimentally in a Wi\nobreakdash-Fi based indoor positioning system, where several regression methods were tested (linear regression, random forest, k-nearest neighbors, and neural networks). The highest positioning error estimation accuracy was achieved for random forest regression, with a mean absolute error of 0.72 m.

Figures (6)

• Figure 1: DAE method concept; a) the algorithm workflow; b) relationship between the positioning result, true user location, and actual and estimated errors
• Figure 2: The process of preparing the DAE training dataset.
• Figure 3: The robotic platform in the experiment environment
• Figure 4: Measurement points' locations. The arrows represent the platform's orientation.
• Figure 5: Empirical cumulative distribution of the signed DAE error