Comparison of Boosting Based Terrain Classification Using Proprioceptive and Exteroceptive Data

The terrain classification is a very important subject to the all-terrain robotics community. The knowledge of the type of terrain allows a rover to deal with its environment more efficiently. The work presented in this paper shows that it is possible to differentiate terrains based on their aspects, using exteroceptive sensors, as well as based on their influence on the rover’s behavior, using proprioceptive sensors. Using a boosting method (AdaBoost), these two sets of classifiers are trained and applied independently. The resulting dual algorithm identifies offline the nature of the terrain on which the vehicle is virtually driving and classifies it according to categories previously labeled, such as sand or grass. Due to the good results obtained for the classification based solely on each type of sensor, this paper concludes that the correlation between data from proprioceptive and exteroceptive sensors could be used for further applications. This paper is a summarized version of the one presented at the ISER conference.