Improving Gaussian Sum Filter to Implement In-motion Yaw Alignment for Quadrotor Attitude Control

This paper presents a new type of integration between the Global Positioning System and Inertial Measurement Unit (GPS/IMU) to provide a higher in-motion yaw alignment accuracy under large initial heading errors when magnetometer data are not reliable. In order to quickly determine the yaw angle without relying on assumptions about the dynamics of the quadrotor, velocities measured from GPS are fused with accelerations and angular velocities measured from the IMU. In contrast to the conventional Gaussian sum filter (GSF), in this paper the forecast weights of the Gaussian mixture model are updated in the time update step of basic estimators, including complementary filters (CF) and extended Kalman filters (EKF). This reduces the number of Gaussian models used in the GSF and speeds up computation on embedded system hardware. © 2023 IEEE.