Unscented Kalman Filter adaptive noise covariance selection for satellite formation flying with Q_leaming

There are many algebraic and dynamic approaches to satellite positioning. Kalman filtering methods are very common in spacecraft position determination. During recent decades, many algorithms and methods based on these filters have been introduced for satellite position estimation. Moreover, due to the coupling between orbital and Attitude dynamics, and also the dependency of some attitude determination algorithms to the precise position of the satellites in formation flying, it is critical to precisely determine the orbit and position of the satellite. This paper adopts the reinforcement learning (RL) method to improve the position estimation The method implemented in this paper for positioning of formation flying satellite is to exploit and merge the Unscented Kalman Filter (UKF) with Reinforcement Learning (RL) to find the optimal value of the process and measurement noise covariance matrices and improve the state estimation of position and velocity of the satellites. The proposed filter is simulated and the results are compared with conventional Kalman filters to demonstrate the effectiveness of this method.