Page 225 - Introduction to Autonomous Mobile Robots

Chapter 5
210
rangefinding, which provides reflectance information in addition to range information.
Choosing a map representation for a particular mobile robot requires, first, understanding
the sensors available on the mobile robot, and, second, understanding the mobile robot’s
functional requirements (e.g., required goal precision and accuracy).

5.5.3 State of the art: current challenges in map representation
The sections above describe major design decisions in regard to map representation
choices. There are, however, fundamental real-world features that mobile robot map repre-
sentations do not yet represent well. These continue to be the subject of open research, and
several such challenges are described below.
The real world is dynamic. As mobile robots come to inhabit the same spaces as humans,
they will encounter moving people, cars, strollers, and the transient obstacles placed and
moved by humans as they go about their activities. This is particularly true when one con-
siders the home environment with which domestic robots will someday need to contend.
The map representations described above do not, in general, have explicit facilities for
identifying and distinguishing between permanent obstacles (e.g., walls, doorways, etc.)
and transient obstacles (e.g., humans, shipping packages, etc.). The current state of the art
in terms of mobile robot sensors is partly to blame for this shortcoming. Although vision
research is rapidly advancing, robust sensors that discriminate between moving animals
and static structures from a moving reference frame are not yet available. Furthermore, esti-
mating the motion vector of transient objects remains a research problem.
Usually, the assumption behind the above map representations is that all objects on the
map are effectively static. Partial success can be achieved by discounting mapped objects
over time. For example, occupancy grid techniques can be more robust to dynamic settings
by introducing temporal discounting, effectively treating transient obstacles as noise. The
more challenging process of map creation is particularly fragile to environmental dynam-
ics; most mapping techniques generally require that the environment be free of moving
objects during the mapping process. One exception to this limitation involves topological
representations. Because precise geometry is not important, transient objects have little
effect on the mapping or localization process, subject to the critical constraint that the tran-
sient objects must not change the topological connectivity of the environment. Still, neither
the occupancy grid representation nor a topological approach is actively recognizing and
representing transient objects as distinct from both sensor error and permanent map fea-
tures.
As vision sensing provides more robust and more informative content regarding the
transience and motion details of objects in the world, mobile roboticists will in time pro-
pose representations that make use of that information. A classic example involves occlu-
sion by human crowds. Museum tour guide robots generally suffer from an extreme amount
of occlusion. If the robot’s sensing suite is located along the robot’s body, then the robot is