This article presents a data fusion method which seeks to obtain better pose estimation of a mobile robot through obtaining a more accurate covariance uncertainty matrix. We seek to compute the state covariance without using the first-order linear approximations of the extended Kalman filter. We consider, unlike standard work done in error propagation and data fusion, the possible correlation between the different sensor pose estimates, odometry and DGPS for the present work, and the autocorrelation of some of the variables involved in the fusion (DGPS data, for the particular case herein presented). Computation of the covariances of each sensor data vector is presented so it takes into account the vehicle kinematics, and hence, its particular characteristics. In order to validate the presented approach, a real outdoor navigation experiment is presented fusing odometry and DGPS data.

Peer Reviewed