Graph-based robust localization and mapping for autonomous mobile robotic navigation

Abstract

Simultaneous Localization and Mapping (SLAM) means to estimate the positions and orientations of the mobile robot and to construct the model of the environment, essential and critical for autonomous navigation and widely used in a large range of application fields, the research goal is to design, implement and validate graph-based robust SLAM algorithm in indoor office-like dynamic scenarios. On the local level, scan matching is executed to estimate the local-relative-roto-translation value: first, pre-processing is performed to filter out the parts corresponding to the moving objects in the raw LIDAR data; second, conditioned-hough-transform-and-linear-regression-based line-segment detection is accomplished to detect the line features from the rest of LIDAR data; third, matching by fitting point to line is applied to estimate the roto-translation value. On the global level, the topological graph is constructed with the previously estimated motion constraints and batch optimization is achieved by a linear solution to estimate the global robot trajectory. Meanwhile, for the local line-feature maps which includes information about the static environment, they are transformed to the global frame based on the robot-pose information and integrated to construct the global-line-feature map. The experiments have verified the effectiveness of this hierarchical algorithm: locally, even when there is much rotation error in the input odometry data, the two sets of laser scan data can still be well matched; globally, the linear solution method can lead to much accurate and efficient results; and the line-feature-based mapping is effective to preserve the key geometrical characteristics of the environment.