A Holistic Approach to Mobile Robot
Navigation using Omnidirectional Vision

Niall Winters
University of Dublin, Trinity College, Ireland. 

Advisors: Prof. José Santos-Victor (ISR/IST), Dr. Gerard Lacey (TCD), Prof. John Byrne (TCD)
Examiners: Prof. James L. Crowley (INRIA, Rhône-Alpes), Dr. Fergal Shevlin (TCD)

Feel free to download a copy in pdf format. All comments welcome via email to: n [dot] winters [at] ioe [dot] ac [dot] uk!


This dissertation presents a novel methodology for indoor, visual-based robot navigation. One of the key observations is that navigation systems should be designed through a holistic approach, encompassing aspects of sensor design, choice of adequate spatial representations with associated global localisation and local control schemes.

We tackle a number of design issues. Taking inspiration from biology, where wide field-of-views are common, we use an omnidirectional camera. This gives us a 360 degree horizontal view of the environment.

An appropriate environmental representation is a key element for successful navigation. We argue that emphasis should be placed on building the appropriate representation rather than relying upon highly accurate information about the environment. Since our robot is designed to travel long distances, we choose a topological environmental representation. The topological map is encoded by a low-dimensional eigenspace obtained from Principal Component Analysis. We detail a local control scheme which allows our robot to effectively use the environmental representation for qualitative localisation.

Finally, we present a method termed, Information Sampling which calculates the most discriminating information within the environment traversed by the mobile robot. By developing a method which allows the robot to focus its attention on this data, it is better able to make effective use of its (limited) computational resources. This enables it to more efficiently handle the complexity of the perception process.

Nedstat Basic - Free web site statistics