Critical Rays Self-adaptive Particle Filtering SLAM

This paper presents CRSPF-SLAM, a critical rays self-adaptive particle filtering occupancy grid based SLAM system that can operate efficiently with different kinds of odometer in real time, in small and large, indoor and outdoor environments for various platforms. Its basic idea is to eliminate the accumulated error of odometer through scan to map matching based on particle filtering. Through some improvements for the original particle filtering method, the lidar system becomes more robust to conduct accurate localization and mapping. Specifically, in our proposed method, particle filter based on Monte-Carlo algorithm is designed to be out-of-step to the odometer; During the scan matching process, the influence of some critical rays selected through a ray-selection algorithm is enhanced and that of the unreliable rays is weaken or removed; The current optimal match value is regarded as the feedback to reset the particle number and the filtering range; Once the optimal pose and scan are obtained, the previous error scan stored in the map will be removed. It is also introduced in the paper that the method can work effectively with dead reckoning, visual odometry and IMU, respectively. And we have tried to use it on different types of platforms — an indoor service robot, a self-driving car and an off-road vehicle. The experiments in a variety of challenging environments, such as bumpy and characterless area, are conducted and analyzed.