Multiple Model Unscented Kalman Smoother for Improved Wind Estimation by a Quadcopter

We propose a multiple model unscented Kalman smoother (MM-UKS) to estimate wind velocities using dynamics of a quadcopter. The algorithm dynamically selects a suitable system model based on the current wind conditions. Additionally, Gaussian Process Regression (GPR) is employed to model the drag force and account for its uncertainty, which enhances the wind estimation accuracy. We present two variants of MM-UKS: interacting and non-interacting MM. Simulation results demonstrate that the proposed MM-UKS method significantly improves the estimation accuracy of wind velocities, outperforming an existing EKF algorithm for quadcopter-based wind estimation. A comparative analysis of the MM-UKS variants shows that the interacting variant better estimates the uncertainty in wind estimation, while the non-interacting variant provides slightly better accuracy.