Comparison of Stochastic Filtering Techniques for GPS/INS Fusion of Sounding Rockets
Eduardo J. de A. J. Filho1; Marcos R. O. A. Maximo1; Júlio C. F. Filho1
1 Aeronautics Institute of Technology
Resumo
This papers considers three different stochastic filtering techniques for the navigation problem of a sounding rocket: extended Kalman filter, unscented Kalman filter and particle filter. Aided navigation is used with global position obtained by low rate GPS devices as well as linear accelerations and angular velocities measured by a low-cost inertial sensor unit with high levels of noise. Flight data is obtained with a high-fidelity six degrees of freedom simulation considering the rocket characteristics of ITA Rocket Design. Euler angles are considered for angular kinematics and they are defined in order to avoid the singularity at 90 degrees of pitch. A comparison is made regarding root mean square error and computational cost. Extended and unscented Kalman filters show better performance in both criteria and the particle filter presents problems related to particle impoverishment. Furthermore, direct roughening technique is used to improve the particle filter.
Palavras-chave: Navigation; GPS/INS Fusion; Sensor Fusion