Magnetic Angular Rate and Gravity Sensor Based Supervised Learning for Positioning Tasks

Abstract

This paper deals with sensor fusion of magnetic, angular rate and gravity sensor (MARG). The main contribution of this paper is the sensor fusion performed by supervised learning, which means parallel processing of the different kinds of measured data and estimating the position in periodic and non-periodic cases. During the learning phase, the position estimated by sensor fusion is compared with position data of a motion capture system. The main challenge is avoiding the error caused by the implicit integral calculation of MARG. There are several filter based signal processing methods for disturbance and noise estimation, which are calculated for each sensor separately. These classical methods can be used for disturbance and noise reduction and extracting hidden information from it as well. This paper examines the different types of noises and proposes a machine learning-based method for calculation of position and orientation directly from nine separate sensors. This method includes the disturbance and noise reduction in addition to sensor fusion. The proposed method was validated by experiments which provided promising results on periodic and translational motion as well.

Bacterial Programming Based Kinematic Chain Estimation of Construction Vehicle

“…IMU raw data was processed with a sensor-fusion algorithm using an iterated EKF approach based on [14], [31], however EKF is not the only solution for orientation calculation. As stated in [33], neural network could also be used to estimate orientation, but in this study EKF was selected, because it is easy to implement and with extra effort, as in [31], it can estimate magnetic disturbance caused errors. Also it provides Earth based absolute angles, which is needed for relative angle calculations.…”