AI for Positioning

This program applies machine learning, deep learning, and probabilistic graphical models to GNSS and multi-sensor positioning. The arc began with supervised classifiers for GNSS multipath/NLOS detection (2017 ITSC; 2018 ANFIS in Journal of Navigation; 2020 Gradient Boosting Decision Tree in Applied Soft Computing) and evolved into deep-learning approaches (LSTM-based GNSS uncertainty prediction; CNN-based environment retrieval in IEEE TIV) and probabilistic graphical models — Factor Graph Optimization — which has become a foundational framework for the field. Most recently the thread reaches into LLM-based data standardization for positioning workflows (Lee, Lin, Hsu, IPIN 2024).

Why it matters. GNSS positioning is a noisy, partially observable, context-dependent inference problem — exactly the kind of problem where modern ML methods extract leverage that hand-crafted models miss. But ML in safety-critical positioning must work with the integrity layer, not bypass it. This is the thread that cross-cuts every other pillar.

Recognition.

• 2024 Most-Cited Paper in NAVIGATION: Journal of the Institute of Navigation — Factor Graph Optimization for GNSS/INS Integration: A Comparison with the Extended Kalman Filter (Wen, Pfeifer, Bai, Hsu)
• IEEE Aerospace and Electronic Systems Magazine review article AI for GNSS (Xu, Hsu et al., 2024)
• Inside GNSS feature article What are the roles of artificial intelligence and machine learning in GNSS positioning? (Hsu, 2020)

Representative publications. AutoW: Self-Supervision Learning for Weighting Estimation in GNSS Positioning (Xu, Hsu, ION GNSS+ 2024); Principal Gaussian Overbound for Heavy-tailed Error Bounding (Yan, Zhong, Hsu, IEEE TAES 2024); ML-based LOS/NLOS Classifier for GNSS Shadow Matching (Xu et al., Satellite Navigation 1(1), 2020).