Attitude Estimation By Separate-Bias Kalman Filter-Based Data Fusion

Abstract

Attitude estimation systems often use two or more different sensors to increase reliability and accuracy. Although gyroscopes do not have problems like limited range, interference, and line of sight obscuration, they suffer from slow drift. On the other hand, inclinometers are drift-free but they are sensitive to transverse accelerations and have slow dynamics. This paper presents an extended Kalman filter (EKF)-based data fusion algorithm which utilizes the complementary noise profiles of these two types of sensors to extend their limits. To avoid complexities of dynamic modelling of the platform and its interaction with the environment, gyro modelling will be used to implement indirect (error state) form of the Kalman filter. The great advantage of this approach is its independence from the structure of the platform and its applicability to any system with a similar set of sensors. Separate bias formulation of the Kalman filter will be used to reduce the computational complexity of the algorithm. In addition, a systematic approach based on wavelet decomposition will be utilized to estimate noise covariances used in the Kalman filter formulation. This approach solves many of the convergence problems encountered in the implementation of EKF due to the choice of covariance matrices. Experimental implementation of the estimator shows the excellent performance of the filter.