Answers are grounded in this paper and cite their sources.
Accurate indoor positioning is a key enabler of smart building intelligence, supporting applications such as occupant localisation, asset tracking, emergency response and energy optimisation. In GPS‐denied environments, WiFi fingerprinting has emerged as a cost‐effective and infrastructure‐ready solution that leverages ubiquitous access points without additional hardware. However, traditional fingerprinting methods face challenges of temporal instability, signal fluctuation and the absence of uncertainty quantification, leading to unreliable performance in dynamic indoor environments. Recent Transformer‐based models have improved feature extraction through attention mechanisms but still struggle to handle noisy or unreliable access points and typically produce overconfident predictions. To address these issues, this paper proposes a Dirichlet‐Enhanced Transformer (DE‐Transformer) architecture that integrates Dirichlet distributions into multi‐head attention to assess the reliability of individual access points whilst generating confidence‐aware position predictions. The proposed framework consists of three modules: WiFi RSSI preprocessing and embedding, a Dirichlet‐enhanced Transformer encoder and an uncertainty‐aware regression decoder. Experimental evaluation demonstrates substantial performance gains—achieving a mean positioning error of 0.90 m and 92.5% accuracy within 1.5 m, outperforming baseline Transformers by 21% in mean error and 53% in worst‐case bounds. Statistical analysis confirms significant improvements with enhanced temporal stability, reduced error variance and calibrated uncertainty estimates. Owing to its robustness, scalability and confidence‐aware inference, the proposed DE‐Transformer represents a promising solution for real‐time, safety‐critical indoor navigation in smart building environments.
No takes yet. Share an insight, caveat, or question.
Nguyen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: