The Computational Intelligence Group of the Faculty of Informatics at the University of the Basque Country (UPV/EHU) in northern Spain have carried out research work in a number of different fields. Ivan Villaverde, a PhD candidate and member of this group, has taken advantage of this multimodality. He used algorithms usually used for for research on the analysis of hyperspectral images and applied them to mobile robots. With this application, the aim is for the robots to enhance their capacity for spatial orientation and their resources for detecting their surroundings -- visual navigation.
Villaverde's PhD thesis is entitled, "On computational intelligence tools for vision-based navigation of mobile robots." In it, he discusses how the visual navigation of small mobile robots can be improved by applying techniques never applied to robotics previously.
His thesis is principally based on an algebraic system that is used in a hyperspectrometric line of research: lattice computing. This involves a system based on series of data (instead of numbers) with concrete internal ordering. As was concluded from the first trials, this technique can be highly useful for enhancing the visual navigation of robots.
Villaverde worked with two types of basic sensors incorporated into mobile robots in order to improve their system of navigation: optic cameras and 3D cameras for range detection. These are the "eyes" of the robot. Villaverde focused on three questions for these "eyes" to see correctly: the location of the robot itself, the capacity of the robot to detect its own movements (being able to fix where it is without access to external information), and the capacity to build a map (distances, obstacles, and so on) of surroundings previously unknown to it.
Thus, the previously mentioned lattice computing (and certain other innovative techniques) was applied to these three questions. Villaverde made use of lattice computing for the self-location of the robot on qualitative maps as well as for the detection of visual markers with optic cameras.
In order to enhance the metric location with 3D cameras, Villaverde applied an innovative hybrid system: combining techniques of evolution and competitive neuronal networks. Evolution techniques are genetic algorithms and neuronal networks are codes that act like the nervous system in humans: Both simulate human mutations and evolution. Villaverde applied these techniques to the 3D cameras and, concretely, to estimating the transformations between 3D views, providing at the same time an estimate of the robot's movement.
He undertook the trials in a shipyard with robots capable of transporting a hose (he chose a shipyard because it has unusual spatial references, thereby providing more testing for detection by the robots). It was a first-phase experiment, so the robots did not have to carry out more than some simple functions, but the thesis showed positive results.
