摘要
This paper presents a modified Rao-Blackwellized Particle Filter (RBPF) approach for the bearing-only monocular SLAM problem. While FastSLAM 2.0 is known to be one of the most computationally efficient SLAM approaches;it is not applicable to certain formulations of the SLAM problem in which some of the states are not explicitly expressed in the measurement equation. This constraint impacts the versatility of the FastSLAM 2.0 in dealing with partially ob-servable systems, especially in dynamic environments where inclusion of higher order but unobservable states such as velocity and acceleration in the filtering process is highly desirable. In this paper, the formulation of an enhanced RBPF-based SLAM with proper sampling and importance weights calculation for resampling distributions is presented. As an example, the new formulation uses the higher order states of the pose of a monocular camera to carry out SLAM for a mobile robot. The results of the experiments on the robot verify the improved performance of the higher order RBPF under low parallax angles conditions.
This paper presents a modified Rao-Blackwellized Particle Filter (RBPF) approach for the bearing-only monocular SLAM problem. While FastSLAM 2.0 is known to be one of the most computationally efficient SLAM approaches;it is not applicable to certain formulations of the SLAM problem in which some of the states are not explicitly expressed in the measurement equation. This constraint impacts the versatility of the FastSLAM 2.0 in dealing with partially ob-servable systems, especially in dynamic environments where inclusion of higher order but unobservable states such as velocity and acceleration in the filtering process is highly desirable. In this paper, the formulation of an enhanced RBPF-based SLAM with proper sampling and importance weights calculation for resampling distributions is presented. As an example, the new formulation uses the higher order states of the pose of a monocular camera to carry out SLAM for a mobile robot. The results of the experiments on the robot verify the improved performance of the higher order RBPF under low parallax angles conditions.
