Generalizing corrective gradient refinement in RBPF for occupancy grid LIDAR SLAM

Abstract

Rao-Backwellized Particle Filter (RBPF) has shown to be a successful framework for Simultaneous Localization and Mapping (SLAM). It is successful because of its non-parametric property in which avoid the local minimum, in turn, excels in the mapping application. However, researches that aim to improve RBPF is declining due to the randomness in the mapping solution and its memory consumption, where the current pervasive approach is the pose-graph SLAM. Recently, Corrective Gradient Refinement (CGR) - a new approach for improving particle filter-based localization - was proposed. In this paper, the traditional RBPF SLAM is augmented with CGR algorithm, and generalized so that it is able to be applied to any kind of robotic sensors. The occupancy grid map structure and LIDAR sensor are used as an implementation case of proposed generalized SLAM algorithm. In the future, this algorithm will be used as a basis for the pose-graph construction.