Support Vector Machine Based Robotic Traversability Prediction with Vision Features

Abstract

This paper presents a novel method on building relationship between the vision features of the terrain images and the terrain traversability which manifests the difficulty of field robot traveling across one terrain. Vision features of the image are extracted based on color and texture. The travesability is labeled with the relative vibration. The support vector machine regression method is adopted to build up the inner relationship between them. In order to avoid the over-learning during training, k-fold method is used and average mean square error is defined as the target minimized to get the optimal parameters based on parameter space grid method. For the traveling smoothness of field robot, the original traversability prediction is transformed to computed traversability prediction based on different initial sub-regions. The optimal path is given by minimizing the sum of computed traversability prediction of all sub-regions in each path. Three experiments are discussed to demonstrate the effectiveness and efficiency of the method mentioned in this paper.