Terrain Traversability Classification Based on SVDD

Terrain travel classification is one of the most important content for robot autonomous navigation, especially in terms of following robots. For this reason, a novel travel ability classification method which is suitable for following robots are studied in this paper. First, slope, roughness, and step height of elevation map are used to describe terrain features. Then, robots will acquire these parameters by means of leader track during following behind, and mark these parameters as terrain travel positive samples. By this, the terrain travel ability classification turns into one-class classification only with positive samples, because the robot may not get the full negative samples and it is dangerous for robots to acquire these samples. At last, the SVDD algorithm is used to train terrain classification model, which enables the robot to have the ability of terrain recognition and learning. The experiment results illustrate the feasibility and effectiveness of the proposed algorithms.