CATEGORIZATION AND CLASSIFICATION OF TERRAIN FOR MOBILE ROBOTS

Abstract

Nowadays, robots are deployed not only indoors but also in outdoor conditions both natural and manmade environments. There are different of types of terrain in the outdoor environment. Some terrains are flat and non-slippery in which the robot can travel easily at relative speed without losing balance, but surfaces can be irregular, grassy or not stiff, so the robot should move or travel on these terrains cautiously. The hazards posed by the terrain are referred to as nongeometric hazards. The mobile robot should be able to determine the type of terrain and plan its movement or motion strategies accordingly. In this thesis, we are considering the problem of terrain classification, in which two approaches are used to identify the terrain that is proprioceptive that is the data input is from sensors attached to the legs and appearance that is the visual images. In proprioceptive-based methods, inputs like vibrations, slips, sinks, and so forth are analyzed and used to learn about the different types of terrain. One of the main drawbacks of proprioceptive-based methods is that robots cannot classify the type of terrain before it traverses it. In appearance-based methods, the inputs are the images of the terrain; this converts the problem into image-processing and classification. Different type of visual sensors is used to provide information on the terrain on which the robot is about to traverse.