Global self-localization of a robot in underground mines

Abstract

We address the problem of navigation by an autonomous agent in underground mines. The robot is equipped with a frontal range sensor, and possesses initial knowledge of the environment's topology in the form of a graph identifying the corridors and their intersections. We describe a simulation test-bed used for our navigation experiments. Then, we define a perception system whose goal is to allow the robot to follow a global path and self-localize in the graph. We introduce an original representation of the range profiles gathered during intersection traversals: the spatio-temporal volume. Using this compact representation, we analyze what the robot can learn about the topology and structure of intersections as it approaches them from corridors.