Building multi-level planar maps integrating LRF, stereo vision and IMU sensors

Building maps of the explored environment during a rescue application is important in order to locate the information acquired through robot sensors. A lot of work has been done on mapping 2D large environments, while the creation of 3D maps is still limited to simple and small environments, due to the costs of 3D sensors and of high computational requirements. In this paper we analyze the problem of building multi-level planar maps. These maps are useful when mapping large indoor environments (e.g., a multi-floor building) and can be effectively created by integrating robustness and efficiency of state-of-the-art 2D SLAM techniques using 2D laser range finder data, with the use of a precise IMU sensor and effective visual odometry techniques based on stereo vision for measuring plane displacements. The main advantages of the proposed solution, with respect to other kinds of 3D maps, are the low-cost of the sensors mounted on the robots and the possibility of exploiting results from 2D SLAM for exploring very large environments. Preliminary experimental results show the effectiveness of the proposed approach.