This paper outlines the methodology and findings of a study aimed at enhancing the performance of the GSHARP PPP algorithm through the application of Machine Learning techniques, specifically targeting the mitigation of multipath effects in GNSS measurements.

The approach uses Machine Learning clustering techniques on key parameters of GNSS measurements to identify and understand patterns in multipath effects. This method serves as an initial step in integrating AI technology into the GSHARP PPP algorithm, with a deliberate choice of lightweight Machine Learning techniques to reduce the integration risk of the outcomes in final implementations. However, more advanced deep learning techniques are not ruled out for future exploration.

Machine Learning algorithms were trained using over 80 hours of data from varied environments like highways, suburban, and urban areas, ensuring their robustness and adaptability in real-world scenarios. The effectiveness of this clustering methodology is evaluated through its integration into the GSHARP PPP algorithm. Different iterations of the PPP algorithm are conducted, incorporating the insights gained from the clustered GNSS measurements in varying ways. The study explores the impact of different clustering strategies on the final positioning solution and the impact of different ways to use the outputs of the clustering on the PPP algorithm.

Preliminary findings suggest a promising enhancement in the robustness and accuracy of GNSS positioning, reducing the maximum errors in several scenarios and showcasing the potential of the introduced Machine Learning techniques. The conclusions drawn from this research not only contribute to advancing GNSS positioning methodologies but also open avenues for future exploration, including the potential integration of more advanced deep learning techniques to further optimize multipath mitigation in high-precision positioning systems.