A decision-tree based multiple-model UKF for attitude estimation using low-cost MEMS MARG sensor arrays. (March 2019)

• Record Type: Journal Article
• Authors: Xu, Xiaolong; Tian, Xincheng; Zhou, Lelai; Li, Yibin
• Is Part Of: Measurement. Volume 135(2019)
• Page Start: 355
• Page End: 367
• Publication Date: 2019-03
• Subjects: Decision tree -- Multiple model -- UKF -- Orientation estimation -- Gradient descent; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2018.11.062
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A decision-tree based unscented Kalman filter for attitude estimation. A set of novel criteria for reliability verification of the sensors. Four different filter models are defined for the UKF. A decision tree to determine which filter model to be utilized by the UKF. Abstract: Micro-electronic-mechanical system (MEMS) is widely used in various applications, especially as a low-cost and small size system for attitude estimation which requires high accuracy and fast response. This work proposes a novel decision-tree based multiple-model unscented Kalman filter (DTMM-UKF) for attitude estimation. It is a quaternion-based attitude estimator that fuses related strap-down magnetic, angular rate, and gravity (MARG) sensor arrays. A set of novel criteria for testing whether the magnetometer and accelerometer are reliable is developed. To improve the anti-interference performance, we define four different filter models for the UKF. Particularly, a decision tree is established to automatically switch filter model based on these reliability test criteria. The priori attitude estimation is obtained from the process model using gyroscope data. Fusing the accelerometer and magnetometer data together, the observation attitude could be solved based on corresponding objective function and Jacobian matrix determined by the filter model. Under the UKF frame, the final optimal attitude could be determined by fusing priori estimation and observed attitude. Experimental tests show that the DTMM-UKF algorithm has better robustness and higher real-time estimation accuracy.