Attitude estimation based on recurrent neural network and vector observations for attitude and heading reference system

Abstract

Attitude and heading reference system (AHRS) is indispensible in miniature unmanned aerial vehicles (UAV).Data fusion for magnetometer, accelerometer, and gyroscope in AHRS is usually implemented using extended Kalman filter (EKF) or complementary filter (CF). But due to the curse of dimensionality, sensor error compensation is difficult to be fully included in the design of EKF or CF. In this paper, a novel attitude estimator based on recurrent neural network (RNN) is introduced. This algorithm takes the observations of gravity vector, geomagnetic vector, and angular velocity vector as its inputs, and it can eliminate sensor errors while implementing dynamic attitude estimation. Simulation and experiment results of the proposed algorithm prove its effectiveness.