Gaussian Process-Based Extended Kalman Filter for Trajectory Estimation in sUAV Traffic Management

This paper introduces the application of Gaussian Process Regression-based Extended Kalman Filter (GPR-EKF) for state estimation. While the traditional EKF is a widely used and efficient method for state estimation in nonlinear systems, it requires an explicit model of the system dynamics. However, when the system dynamics are unknown or only partially known, the EKF is no longer an appropriate choice. This study presents the GPR-EKF as an alternative for state estimation when system dynamics are unknown, but measurements are available from onboard sensors. In the proposed framework, a Gaussian Process Regression-based propagation model is utilized, trained on observed input-output data to predict the mean and covariance of the states. The states of the dynamical system under consideration are correlated; therefore, a Multi-Output Gaussian Process (MOGP) is implemented to capture the correlation. The proposed GPR-EKF is then employed to assess collision risk for sUAV traffic management.