Adaptive invariant Kalman filtering for attitude estimation on <i>SO</i>(3) thorough feedback calibration of prior error covariance

作     者：Wang, Jiaolong Li, Minzhe 

作者机构：Jiangnan Univ Inst Automat Minist Educ Key Lab Adv Proc Control Light Ind Wuxi Jiangsu Peoples R China Shanghai Jiao Tong Univ Sch Aeronaut & Astronaut Shanghai Peoples R China 

出 版 物：《IET CONTROL THEORY AND APPLICATIONS》 (IET控制论与应用)

年 卷 期：2021年第15卷第14期

页      面：1906-1914页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0804[工学-仪器科学与技术] 0811[工学-控制科学与工程] 

基　　金：National Natural Science Foundation ofChina (NSFC) Fundamental Research Funds for theCentral Universities [JUSRP121022] 

主　　题：Signal processing theory system process noise stochastic processes Other topics in statistics stochastic systems Kalman filters state-space methods geometry property statistics parameter AIKF Combinatorial mathematics calibration Time-varying control systems invariant attitude dynamics stochastic feedback-based covariance calibration scheme attitude estimation problems covariance matrices Lie groups feedback Algebra feedback calibration adaptive invariant kalman filter Filtering methods in signal processing covariance propagation step unknown process noise statistics error covariance geometry matrix Lie groups feedback stochastic sequence attitude control 

摘      要：For invariant attitude dynamics evolving on matrix Lie groups, by proposing the stochastic feedback-based covariance calibration scheme, an adaptive invariant Kalman filter (AIKF) is elaborated to deal with the attitude estimation problems corrupted by unknown or inaccurate process noise statistics. The invariant Kalman filter (IKF) takes into account the geometry property of attitude dynamics and can boost the estimation performance;however, IKF requires accurate knowledge of the noise statistics and an incorrect noise parameter is prone to deteriorating the precision of final estimates. To eliminate this impact, instead of using the original covariance propagation step of IKF, the prior error covariance of the proposed AIKF is online calibrated based on the posterior information of the feedback stochastic sequence. As the main advantage, the statistics parameter of system process noise is no longer required in the proposed AIKF and the negative influence by unknown/incorrect noise parameters can be reduced significantly. The mathematical foundation for the new adaption scheme of AIKF is also presented. The AIKF s advantage in filtering adaptability and simplicity is further demonstrated by numerical simulations.