Terrain Classification for Mobile Robots on the Basis of Support Vector Data Description

The ability to detect traversable terrains is essential for autonomous mobile robots to guarantee safe navigation. In this paper, we present a method for terrain classification for wheeled mobile robots. Our scope is limited to mobile service robots that are used for surveillance or delivery in semi-structured urban environments. A reliable terrain detection scheme is required for both indoor and outdoor applications anytime. A low-cost Lidar (Light detection and ranging) is adopted for terrain detection. To deal with intrinsic measurement errors and uncertainties of the Lidar, the classification criteria are trained through a supervised learning approach. Training data are obtained from manual driving at target environments. Various decision boundaries resulted from a variety of floor conditions, sensor types and robot platforms. The proposed terrain classification scheme is experimentally tested in success.