Unscented Kalman Filtering for Attitude Determination Using MEMS Sensors

This paper presents the results of a quaternion-based unscented Kalman filtering for attitude estimation using low cost MEMS sensors. The unscented Kalman filter uses the pitch and roll angles computed from gravity force decomposition as the measurement for the filter. The immeasurable gravity accelerations are deduced from the outputs of the three axes accelerometers, the relative accelerations, and the accelerations due to body rotation. The constraint of the four elements of the quaternion method is treated as a perfect measurement and is integrated into the system to form a constrained unscented Kalman filter. The heading angle is obtained from a complimentary filter which uses the heading signal derived from the magnetic force information from an electronic magnetic sensor and the GPS-derived heading as the inputs. An experiment using an in-house designed motion platform is conducted to evaluate the proposed algorithm. The noise characteristics of the sensor signals are examined using the laboratory data. Approximations of the time-varying noise variances of the measured signals are obtained through Taylor series expansions. The algorithm is intuitive and easy to implement. Moreover, the proposed algorithm and the filter design are successfully demonstrated through a complete set of flight test data.