Abstract: Full-attitude determination methods are needed in many fields, such as UAV, robot, and human motion capture. Using MEMS inertial/magnetic sensors to determine the attitude of a object is a low cost, high performance, and commonly used scheme in these fields. To alleviate the problem that the magnetometer is susceptible to external ferromagnetic interference, the paper proposes a two-stage quaternion-based attitude update method: in the first stage, a cost function is constructed by establishing the gyroscope and accelerometer measurement models, which can be optimized by gradient descent to update the pitch and roll angles; in the second stage, a cost function is constructed by using the magnetometer measurement model, which can be optimized by gradient descent to update the yaw angle. The posteriori estimated quaternion is used to calculate the estimated ferromagnetic interference is regarded as the priori ferromagnetic interference at the next step. The experimental result shows that the proposed method can be used to effectively correct the cumulative error caused by angular rate integration and improve the attitude determination accuracy, and has the capability of anti-ferromagnetic interference, which can improve the environmental adaptability of the attitude determination system. Key

Key words: microinertialandmagneticmeasurementunit, attitudedetermination, double-stageattitudeupdate, gradientdescentmethod, ferromagneticinterferencecompensation