Terrain Typing for Real Robots

Abstract

Many robotics tasks require an ability to determine quickly the nature of the terrain surrounding the robot. While much attention has been given to the general problem of terrain typing, the problem of effective real-time terrain typing remains open. For robot missions such as construction site work, military reconnaissance, hazardous waste removal, and planetary exploration this problem must be addressed. In particular, for cross country navigation with a wheeled vehicle, the robot needs to know where the vegetation is and where the rigid obstacles are because frequently the optimal, if not the only, path will pass through vegetation. Our groups have independently researched the problem of finding vegetation in a scene, and have developed systems tuned to the specific demands of real-time terrain typing for robots. This paper looks at three classifiers of increasing dimensionality and describes their applicability to different aspects of the terrain typing problem.