Page 231 - Innovations in Intelligent Machines

224 J. Gaspar et al.
A critical component of any perceptual system, human or artificial, is the
sensing modality used to obtain information about the environment. In the
biological world, for example, one striking observation is the diversity of ocular
geometries. The majority of insects and arthropods benefit from a wide field
of view and their eyes have a space- variant resolution. To some extent, the
perceptual capabilities of these animals can be explained by their specially
adapted eye geometries. Similarly, in this work, we explore the advantages
of having large fields of view by using an omnidirectional camera with a 360
degree azimuthal field of view.
Part of the power of our approach comes from the way we construct rep-
resentations of the world. Our internal environmental representations are tai-
lored to each navigation task, in line with the information perceived from the
environment. This is supported by evidence from the biological world, where
many animals make alternate use of landmark-based navigation and (approxi-
mate) route integration methods [87]. Taking a human example when walking
along a city avenue, it is sufficient to know our position to within an accu-
racy of one block. However, when entering our hall door we require much
more precise movements. In a similar manner, when our robot is required to
travel long distances, an appearance-based environmental representation is
used to perceive the world [89]. This is a long-distance/low-precision naviga-
tion modality. For precise tasks, such as docking or door traversal, perception
switches from the appearance-based method to one that relies on image fea-
tures and is highly accurate. We characterize these two modes of operation
as: Topological Navigation and Visual Path Following, respectively.
Combining long-distance/low-precision and short-distance/high-accuracy
perception modules plays an important role in finding efficient and robust
solutions to the robot navigation problem. This distinction is often overlooked,
with emphasis being placed on the construction of world models, rather than
concentrating on how these models can be used effectively.
In order to effectively navigate using the above representations, the robot
needs to be provided with a destination. We have developed human-robot
interfaces for this task using (omnidirectional) images for interactive scene
modelling. From a perception perspective, our aim is to design an inter-
face where an intuitive link exists between how the user perceives the world
and how they control the robot. We achieve this by generating a rich scene
description of a remote location. The user is free to rotate and translate this
model to specify a particular destination to the robot. Scene modelling, from a
single omnidirectional image, is possible with limited user input in the form of
co-linearity, co-planarity and orthogonality properties. While humans have an
immediate qualitative understanding of the scene encompassing co-planarity
and co-linearity properties of a number of points in the scene, robots equipped
with an omnidirectional camera can take precise azimuthal and elevation
measurements.