Robust Visible Light Positioning: a Comparative Study of Machine Learning Estimators in Spatial Segmentation and Transmitter Failure Scenarios

Visible Light Positioning (VLP) systems present a promising alternative for indoor positioning by leveraging existing Light-Emitting Diode (LED) lighting infrastructure to achieve high accuracy. However, challenges such as uneven error distribution and vulnerability to transmitter failures can compromise system robustness. This study conducts a comprehensive comparative analysis of machine learning estimators for VLP, emphasizing spatial segmentation and transmitter failure scenarios. An experimental setup with four LED transmitters was utilized to collect a dataset of received signal strengths across various positions within a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1.2 \times 1.2 \mathrm{m}$</tex> testbed. Polynomial interpolation was applied to augment the dataset, expanding the number of samples from 169 to 625. The workspace was divided into center and edge regions based on cumulative angles of emission, and classification models were employed to validate this segmentation. Several machine learning estimators were evaluated for position estimation. Gaussian Process (GP) regressor achieved the lowest overall Root Mean Square Error (RMSE) of 1.973 cm, particularly excelling in the center region. All models exhibited lower errors in the center compared to the edge, highlighting significant regional disparities. Robustness under transmitter failure was assessed by simulating the failure of one transmitter. GP regressor demonstrated resilience, maintaining low RMSE and experiencing moderate performance degradation (28.1 %). These findings underscore the importance of spatial segmentation in VLP systems and the effectiveness of GP and Support Vector Machine estimators in achieving robust and accurate positioning. The study suggests that accounting for regional differences and potential transmitter failures can lead to more resilient and precise VLP solutions.