Flight Attitude Estimation Of Quadrotor UAV Based On GRU

Attitude estimation is the premise to ensure stable and controllable flight for unmanned aerial vehicles(UAV),and thus attitude and heading reference systems(AHRS)are widely used in quadrotor UAVs’ flight control.At present,attitude estimation for quadrotor is usually based on the combination of magnetometer,accelerometer,and gyroscope(MARG).But MARG sensor can be easily affected by various disturbances,such as vibration,external magnetic interference,gyro drift,etc.Such disturbances can lead to large errors in attitude measurement by single sensor.In addition,due to the inherent noise of low-cost MARG sensors,there is a significant nonlinear correlation between attitude error and noise,which is difficult to be mathematically modeled.Therefore,it is of great research value and significance to select an appropriate attitude estimation algorithm to effectively fuse the measured values of multiple sensors to achieve high-precision attitude estimation.In order to fit the nonlinear relationship between attitude error and sensor output,gated recurrent neural network(GRU)algorithm is used for multi-sensor data fusion.This research mainly focuses on GRU-based multi-sensor fusion solutions for attitude estimation.To improve the accuracy of attitude estimation,the following issues are discussed.(1)Firstly,the theories of quadrotor attitude measurement,MARG sensor’s principle and error sources are introduced.(2)Attitude estimation algorithm based on GRU neural network accelerometer and magnetometer is proposed.The proposed algorithm is verified by simulation,and the results prove its feasibility.(3)Accelerometer and magnetometer have poor dynamic responses when quadrotor experiences large motion.To solve this problem GRU neural network algorithm is used to implement data fusion of MARG sensor.Since gyroscope has better dynamic response,it can help to improve the dynamic accuracy.Experimental results show that the proposed method can achieve multi-sensor fusion and make full use of the characteristics of gyroscope,so as to improve the dynamic accuracy of quadrotor.(4)Optical flow sensor can extract motion information from image sequence.But the major problem is that the optical flow can be caused by both translational and rotational movements,which are difficult to be distinguished from each other.To solve this problem,the GRU neural network algorithm is used to extract the attitude information in the optical flow measurement and fuse it with the MARG sensor to improve the attitude estimation accuracy of the quadrotor.The experimental results show that the algorithm can use the attitude information in optical flow measurement to further improve the 3D attitude accuracy.Finally,the three proposed attitude estimation algorithms are compared with the commonly used extended Kalman filter(EKF)algorithm.The experimental data show that the proposed attitude estimation methods have higher attitude accuracy in quadrotor flight experiments.