A UAV attitude fusion algorithm based on gradient descent-improved EKF

When the UAV is in accelerated motion, the non-gravitational acceleration and body vibration lead to high-frequency errors in the accelerometer acquisition data, which affects the accuracy of the attitude calculation of the UAV. To solve the above problems, a UAV attitude solution fusion algorithm based on gradient descent-improved Extended Kalman Filter (EKF) was designed. Firstly, the gradient descent method is used to solve the sensor data to reduce the error caused by the accelerometer and magnetometer. Secondly, the noise covariance matrix and the non-gravity acceleration function are established to improve the shortcomings of the traditional noise matrix that cannot suppress the interference of non-gravity acceleration. Finally, the gradient descent solution value is used as the measured value of the improved EKF to improve the accuracy and filtering effect of the measured value and realize the attitude update. The experimental results show that the maximum error of the UAV roll angle is 0.017 degrees in the static experiment. Compared with the complementary filtering and EKF algorithm, the error is reduced by 82.65% and 57.50%. In the dynamic experiment, compared with the complementary filtering algorithm, the fusion algorithm error is reduced by at least 72.75% in pitch angle; the roll angle reduced by at least 54.84%; the yaw angle reduced by at least 59.09%. Compared with the EKF algorithm, the fusion algorithm error is reduced by at least 65.33% in pitch angle; the roll angle reduced by at least 27.97%; the yaw angle reduced by at least 40.00%. The UAV attitude fusion algorithm based on gradient descent-improved EKF can effectively suppress the influencing factors such as non-gravitational acceleration interference and high-frequency vibration of the airframe, and improve the attitude-solving accuracy of the UAV.