Smartwatch PDR/GNSS Integrated Positioning Using MC-EKF with Adaptive Covariance under Measurement Uncertainty

This paper presents a robust fusion framework for smartwatch-based pedestrian navigation by integrating inertial Pedestrian Dead Reckoning (PDR) and Global Navigation Satellite System (GNSS) data. Although GNSS provides absolute positioning, experiments revealed that transient errors may occur even under open-sky conditions where HDOP remains nearly constant. As a result, conventional EKF fusion often produces unstable updates. To address this issue, a Maximum Correntropy Criterion Extended Kalman Filter (MC-EKF) is employed. The PDR module predicts motion using stride-length and heading estimation, and its process noise is modeled from their uncertainties. During GNSS updates, the MC-EKF adaptively adjusts the measurement and error covariance using residual statistics to mitigate unreliable observations. Real-world experiments with a smartwatch demonstrate improved trajectory stability and lower horizontal RMSE compared to conventional EKF fusion, validating the robustness of the proposed method.