A Novel Hybrid Attitude Fusion Method Based on LSTM Neural Network for Unmanned Aerial Vehicle

Unmanned aerial vehicle (UAV) is a highly coupled and multivariable nonlinear complex system, the attitude fusion of which is often disturbed by the noise of low-cost inertial measurement unit (IMU). In order to improve the UAV attitude fusion precision, a novel hybrid attitude fusion method is introduced in this paper. An error model between the IMU output and attitude is established by a long short term memory neural network (LSTM NN). The attitude fusion error can be estimated well by using the powerful nonlinear fitting and time series processing ability of the LSTM. The experimental results demonstrate that the root mean square error (RMSE) of pitch, roll and yaw can be reduced to 0.69 degrees, 0.73 degrees and 0.59 degrees respectively, which effectively improve the accuracy of attitude fusion.