Experimental comparison of EKF and Constraint Manifold particle Filter for robot localization

This communication presents the comparison of two data fusion methods : the extended Kalman filter and the constraint manifold particle filter applied to the problem of vehicle positioning estimation. The comparison is made with reference to precise positioning data given by a differential GPS. The estimation fuses natural GPS data of metric precision with proprioceptive data from two odometers. Both filters have been implemented onboard of an experimental vehicle and tested in real time. The results obtained on one of the numerous experimented trajectories is presented as an example and analyzed. The paper mainly aimed at describing the comparison framework and at reporting the large amount of experimentations realized.