Towards geometric mapping for semi-autonomous mobile robots

Semi-autonomous mobile robots are a promising alternative for tasks that are too challenging for autonomous robots. Especially in an unstructured environment, full autonomy is still far from being realized. In order to enable the human operator to control the robot properly, visualization of the environment is crucial. In this paper, we introduce a pipeline for geometric mapping that uses narrow field of view RGB-D cameras as input source and builds a geometric map of the environment while the robot either is operated manually or moves autonomously. Geometric shapes are extracted from subsequent sensor frames and are clipped and merged in a geometric feature map. Evaluation is done both in simulation and on the real robot.