An Attitude Compensation Method based on Neural Network Using Data from MEMS MARG Sensors

Abstract

Aiming at the attitude estimation errors existed in the attitude measurement system using the magnetic, angular rate and gravity (MARG) sensors, a novel attitude compensation algorithm based on the neural network using data from sensor networks is proposed in this paper. It provides a more accurate measurement with simple implementation. A simple Kalman filter (KF) is designed to achieve prior attitude estimation. A back-propagation neural network is designed to compensate the attitude errors for the KF. The neural network is trained by the data from MARG sensor networks and errors of attitude estimation algorithm. Taking the data from triaxial MARG sensors as inputs, the trained network could predict the estimated attitude errors in Euler-angles form. Moreover, to validate the effectiveness of the proposed algorithm, two different experiments are accomplished. In the first experiment, the designed neural network is utilized to compensate the attitude estimation errors of a micro quadrotor helicopter using the publicly available datasets. In the second experiment, the KF is combined with the neural network to estimate the attitude of a self-designed single axis platform. Results show that the estimated attitude is closer to actual attitude after compensation which indicates that it is effective to utilize the neural network to compensate for estimation errors in attitude detection field even the attitude estimation algorithm is a simple Kalman filter.