Extended Kalman Filter tuning in attitude estimation from inertial and magnetic field measurements

The extended Kalman filter (EKF) has been the workhorse of real time attitude estimation problems, for several years now. However, an essential and unsolved issue in the practical implementation of the EKF is the selection of the process and measurement noise covariance matrices. In this article, we evaluate experimentally an estimation algorithm that solves a gyro free quaternion formulation of Wahba´s problem. This algorithm is based on an EKF and a least squares algorithm sensor fusion procedure. In particular, we address the tuning issues of the covariance matrices in the EKF and the stop criteria and the initial condition in the sensor fusion procedure. Unfortunately, our experimental results show that the algorithm fine tuning is not an easy task and our best results, by the time being, rely on gyroscopic measurements.