Vibration-based Terrain Classification for Mobile Robots Using Support Vector Machine

In order to enhance terrain classification accuracy of mobile robots, a feature extraction method is proposed based on time amplitude domain analysis, and a new voting decisions classification algorithm is proposed to deal with the situation of same number of votes via one-against-one support vector machine (SVM) program in LIBSVM. A four-wheeled mobile robot on which arm accelerometers in x,y,z directions and a microphone in z direction are installed in left front wheel, is used to get the acceleration and sound pressure signals of wheel-terrain interaction by traversing on sand, gravel, grass, soil and asphalt terrains with six different velocities respectively. Five kinds of terrains in each velocity are classified by the proposed algorithm, and the average classification accuracy is 88.7%.