Mobile Robot Global Localization Using Unique Objects in Indoor Scenes

Abstract

Localization or pose estimation is a fundamental capability for an autonomous mobile robot, especially in navigation tasks. With a known occupancy grid map, wheel encoder data, laser LiDAR data, and a particle filter localization framework, the mobile robot can localize itself in most indoor scenarios. However, the mobile robot will fail to localize when the environment has similar areas or few geometric structural features. To solve the problem, we propose to use an RGB-D camera to detect the unique object in each sub-region as a reference which is useful for assisting the localization task. The semantic object information will be associated with the grid map coordinate in the mapping stage. In the localization phase, the mobile robot moves to a suitable area to detect the unique object, then completes the global localization task based on the known object pose. The conducted experiment shows that the proposed method is effective while traditional laser-based particle filter always fails.