Robot Localization in Complex Home Environments Based on Layered Sub-Maps and Observation Residuals

Home environments are complex and full of objects, which presents challenges for service robots trying to navigate and localize themselves. To tackle this issue, this paper proposes a method for robot localization in complex home scenarios based on layered sub-maps and observation residuals. Firstly, a 3D point cloud of the indoor scene is synthesized, and longitudinal point cloud filtering is applied to project the obtained layered point cloud into corresponding virtual lasers and layered sub-maps. Next, a map selection method based on edge and corner detection is devised to choose localized layered sub-maps. Then, the Adaptive Monte Carlo Localization (AMCL) algorithm is independently executed on each map, yielding localization results for the layered sub-maps. Finally, a similarity evaluation algorithm based on observation residuals is developed to determine the optimal robot pose. This involves using Extended Kalman Filtering (EKF) to estimate and eliminate outliers from odometry and IMU data, along with comparing the residuals of the robot-observed point cloud virtual laser with the map virtual laser to select the best localization result. Experimental results demonstrate that our method achieves more accurate localization in complex home environments.