Terrain field SLAM and uncertainty mapping using Gaussian process

This paper presents a method to infer the terrain field and robot position exploiting the vibration obtained from the interaction between terrain and mobile robot. In order for robot localization and mapping in unknown area, simultaneous localization and mapping (SLAM) technique is applied to map the surrounding area. In this case, when SLAM is performed using a field such as WiFi, magnetic signal or terrain, a complete field map should be inferred for the unexplored region as well as the explored one. Also the uncertainty should be indicated for the inferred field, so that the field map can be used as observation model for SLAM. Therefore, by modeling the observation model for the terrain field with the Gaussian process, we estimate the observation probability distribution for the unknown regions. The inferred observation distribution can be used not only by field maps, but also by efficient path planning. We demonstrate the proposed method with the odometry of mobile robot navigating the testbed, and observations of terrain feature using simulation.