A new robust quaternion-based initial alignment algorithm for stationary strapdown inertial navigation systems

Abstract

A new robust quaternion Kalman filter is developed for accurate alignment of stationary strapdown inertial navigation system. Most fine alignment algorithms have tried to estimate the biases of gyroscopes and accelerometers to reduce the errors of the alignment process. In stationary platforms, due to fixed inputs for sensors, the summation of various errors such as fixed bias, misalignment, scale factor, and nonlinear errors acts like one bias error, and then the identification of each error will be impossible. The observability of gyros and accelerometers’ biases has also been studied. But, nowadays, we know that all of these unknown parameters are not observable. Then this problem can increase the complication of the alignment algorithm. The accelerometers’ errors mainly affect the errors of the roll and pitch angles, but a big portion of the heading’s error results from the gyroscopes’ errors. Modeling of all errors as additional states without considering the observability parameters has no benefits, but will increase the filter’s dimension, so the filter’s performance will decrease. In this study, due to the observability problem, a new robust multiplicative quaternion Kalman filter is designed for the alignment of a stationary platform. The presented algorithm does not estimate the sensors’ errors, but it is robust to uncertainty in the sensors’ errors. In the proposed scheme, the bounds of parameters’ errors are introduced to filter, and the filter tries to remain robust with respect to these uncertainties. The method uses the benefits of quaternions in attitude modeling, and then the robust filter is adapted to work with quaternions. The ability of the new algorithm is evaluated with MATLAB simulations. The outcomes show that the presented algorithm is more accurate than other traditional methods. The extended Kalman filter with accelerometers’ outputs and the horizontal velocities as the measurement equations and additive quaternion Kalman filter are used for comparisons.