Reducing Training Data for Indoor Positioning through Physics-Informed Neural Networks

In this work, we propose a novel framework based on Physics-Informed Neural Networks (PINNs) for directly estimating indoor positions, a method that, to the best of our knowledge, has not been previously explored. Training is performed on a public BLE dataset that includes a variety of indoor scenarios, including Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) conditions caused by human body signal attenuation. The integration of physics-compliant synthetic data during the training phase significantly reduces dependence on large-scale real-world datasets, enabling the use of a simple Multilayer Perceptron (MLP) architecture. Our results demonstrate that combining PINNs with real-world measurements enhances model generalization without compromising accuracy.