Multisensor Fusion SLAM Research Based on Improved RBPF-SLAM Algorithm

Abstract

Simultaneous localization and map construction (SLAM) technology provides the foundation for indoor robots to realize autonomous path planning. The Rao-Blackwellized particle filtering (RBPF) algorithm is widely used to obtain information and perform map construction in unknown environments. This paper proposes a multisensor fusion algorithm that improves the RBPF-SLAM algorithm by addressing the issues of particle distribution errors and degradation. To achieve this, the extended Kalman filter (EKF) is first adopted to effectively fuse odometry and inertial navigation (inertial measurement unit (IMU)) data as the initial positional information. Then, a high-precision light detection and ranging (LIDAR) observation model is integrated to calculate the proposed distribution, and a threshold is introduced to determine the number of valid particles to simplify the resampling step. Finally, raster map construction experiments were carried out on both the Gazebo simulation environment and Robot Operating System (ROS)-based mobile robot Keepbot platform in different scenarios. The experimental results demonstrate that the optimized RBPF-SLAM algorithm generates a clearer and more complete map outline, which can effectively improve the accuracy of robot pose estimation, acquire a reliable 2D raster map with a reduced number of particles, and greatly reduces the computational effort.