Vision-only mobile robot navigation with topological maps
Toon Goedemé, Luc Van Gool · Lirias (KU Leuven) · 2009
© 2009 by Nova Science Publishers, Inc. All rights reserved. When humans navigate, they use mainly their eyes as input from the environment. Indeed, in a new place, we look around and remember the appearance of each place. We automatically organise this visual information as a kind of visual map, which can be used to recognise places later on and to find the way to a certain goal. These tasks are known respectively as localisation and path planning in mobile robotics. Experi- ments show that the mental map of human navigators is organised topologically, rather than metrical. In this work we present a total navigation system for autonomous mobile robot navigation based on these insights. With only an omnidirectional camera as sensor, this system is able to build automatically and robustly accurate topologically organ- ised environment maps of a complex, natural environment. It can localise itself using that map at each moment, both at startup phase (kidnapped robot) and if knowledge of former localisations is present. The topological nature of the map, similar to the intu- itive maps humans use, is memory-efficient and enables fast and simple path planning towards a specified goal. We developed a real-time visual servoing technique to steer the system along the computed path. The power of this approach lies in the combination of a topological world model with state-of-the-art image matching techniques based on fast wide baseline local fea- tures. Techniques as SIFT and, more recently, SURF make it possible to find image correspondences without the need for artificial markers in the scene, and also enables an efficient description of the environments. Moreover, it enables the computation of a visual similarity measure, useful to organise visual databases in a topological world model. We applied our navigation approach on autonomous wheelchair navigation of which this text presents experimental results.