Estimation of Robot Motion State Based on Improved Gaussian Mixture Model

For robot motion state estimation accuracy under complex environment to improve the problem, an improved Gaussian summation cubature Kalman filter is proposed for a kind of nonlinear non-Gaussian system by improving expectation-maximum algorithm and Gauassian merging method. Firstly, the weighted information is introduced to help improve the penalty item of the objective function in the expectation-maximum algorithm, so that more comprehensive parameter information can be considered in the optimization process to achieve the purpose of reducing the number of iterations of the EM algorithm and increasing the convergence speed. Then, based on the Gaussian merging methods using the Mahalanobis distance and the Kullback-Leibler distance, respectively, one fusion mode that can effectively combine the two types of Gaussian merging method is proposed. The Mahalanobis distance and the Kullback-Leibler distance are used to merge the Gaussian mixture items separately, and then the obtained Gaussian mixture items are weighted and fused to improve the merging performance and fidelity of multiple Gaussian items in the Gaussian sum filtering. Finally, the Gaussian sum cubature Kalman filter framework of nonlinear non-Gaussian system is applied to estimate the motion state of the robot in complex environment. Theoretical analysis and simulation results show that the new method can achieve better state estimation accuracy for robot motion and obtain stronger robust performance. © 2022 Science Press. All rights reserved.