Image based navigation using a topological map

Abstract

A new approach on robot navigation is described which enables the robot to drive through the environment using only a vision system as its sensor. In an exploration phase the robot is driven manually through the environment while taking images. By computing similarities between these images the robot constructs a topological map in the form of an appearance graph. Navigation on this graph involves homing from one node to the other until the goal node is reached. No explicit metrical information is modelled by this approach. Rather the distances in the topological map provide directly information about the ability and the robustness of visual navigation in the environment. Real world experiments were conducted under varying environmental conditions. 1