An Approach to 3D SLAM for a Mobile Robot in Unknown Indoor Environment towards Service Operation

In this paper, an approach of 3D SLAM (Simultaneous Localization and Mapping) is presented for mobile robot towards service operation. The scheme can be briefly concluded as follows. Firstly, various robust and stable point and line features are extracted from RGB images and their corresponding 3D information is obtained from the depth image as landmarks for robot localization. Secondly, considering the occlusion and segmentation errors of line features in RGB images, a class of optimal decision models are constructed with a kind of Huber kernel function of point errors and line errors so as to find the optimizing solutions of rotation matrix and translation vector for the pose estimation of service robot. Thirdly, a pose graph is generated as an effective reference for localization along with the selection of keyframes and loop closure based on a bag-of-words model. Moreover, the pose graph is optimized using bundle adjustment algorithm and the robot trajectory is also globally optimized to correct drift error and enhance the consistency of map. Along with the localization of robot, a 3D sparse nap and a point cloud map is built simultaneously using the features in keyframes. A series of simulation experiments are carried out with public datasets to verify the performance advantages of our proposed approach.