Soft Computing for Visual Terrain Perception and Traversability Assessment by Planetary Robotic Systems

Abstract

This paper discusses technical challenges and navigational skill requirements of mobile robots for traversable path planning in natural environments similar to Mars surface terrains. Different methods for detecting salient terrain features based on imaging texture analysis techniques are described. In particular, three competing soft computing techniques are presented for terrain traversability assessment: a rule-based terrain classifier, a neural network-based terrain classifier, and a fuzzy-logic terrain classifier. Each terrain classifier divides a region of natural terrain into finite sub-terrain regions and classifies terrain condition exclusively within each sub-terrain region based on terrain visual clues. Image processing techniques are applied for aggregative fusion of sub-terrain assessment results. Results of a comparative performance evaluation of all three terrain classifiers are presented. The last two terrain classifiers are shown to have remarkable capability for traversability assessment, which facilitates navigation in unstructured natural terrain environments.