Acta Geodaetica et Cartographica Sinica ›› 2022, Vol. 51 ›› Issue (2): 212-223.doi: 10.11947/j.AGCS.2022.20210082

• Photogrammetry and Remote Sensing • Previous Articles     Next Articles

Advanced quaternion unscented Kalman filter based on SLAM of mobile robot pose estimation

ZHAO Leyang, YAN Li   

  1. School of Geodesy and Geomatics, Wuhan University, Wuhan 430079, China
  • Received:2021-02-15 Revised:2021-11-10 Published:2022-02-28
  • Supported by:
    The National Key Research and Development Program of China (No. 2020YFD1100203)

Abstract: In automatic motion controlled mobile robot system, the estimation and correction of its own pose is very crucial for the motion of robot. Kalman filter is a classical method to solve the problem of simultaneous localization and mapping (SLAM) in robot system. Compared with Kalman filter, unscented Kalman filter (UKF) uses nonlinear model directly, avoids operation of Jacobian matrix of complex nonlinear function. In this paper, based on the unscented Kalman filter, sigma points are selected by square root decomposition of prior covariance, and then weighted mean and covariance are calculated. In addition, Quaternion is used to represent attitude of robot, and quaternion vector is converted to rotation space for matrix operation. According to the characteristic of square root decomposition and quaternion vector, a quaternion square root unscented Kalman filter is proposed. By comparing the robot poses estimation results on quaternion square root unscented Kalman filter (QSR-UKF), square root unscented Kalman filter (SR-UKF) and extended Kalman filter (EKF), the simulation results show that the proposed QSR-UKF method is effective.

Key words: mobile robot, simultaneous localization and mapping, unscented Kalman filter, quaternion

CLC Number: