Abstract
A fusion algorithm for attitude estimation of MAV(Micro Aerial Vehicle) is presented. The algorithm use quaternion to represent rotation. Attitude quaternion errors and drift bias of gyros are selected to construct a Kalman filter state vector. The output of GPS is used to correct the accelerometer readings. The local earth magnetic and gravity are used to compute observation of Kalman filter by Levenberg-Marquardt algorithm. The algorithm can converge quickly to rotation attitude of small flight vehicle.