Summary
|
For mobile robotics, the task of navigation is essential. Navigation
comprises selflocalization, planning, and motion. Both self-localization
and planning require a representation of the environment, e.g. a metric
or topological map. A common framework representing both topological and
metric knowledge is the route graph. The proposed project will investigate
the integration of spatial knowledge for robot navigation following this
approach, especially its potential as a cognitively adequate representation
for human-machine-interaction in the navigation domain. Robots are not
able to recognize the complete environment; they only perceive a very
limited part of it. Using a laser range finder, for example, they only
determine the distances to obstacles at a certain height, i.e., they cannot
detect objects at other heights, cannot see holes in the ground, etc.
Similar constraints apply to the use of existing maps, because they do
not contain all objects that may be perceived by the sensors of a robot,
e.g. furniture is missing in most maps. In addition, the information
perceived by a robot or extracted from a CAD-representation may be
inadequate to communicate with human users, because neither a topological
nor a metric map is easy to understand without named nodes, edges, or
rooms. A route graph may contain this information, but a central question
is how it can be acquired and how it will be represented. The project has
two goals: on the one hand, pre-existing maps will be semiautomatically
adapted to robot navigation, i.e. converted into route graphs, easing the
process of map generation and maintenance for robotic applications. On the
other hand, route graphs will be augmented by qualitative spatial information
resulting from a reduced set of natural language expressions given by an
instructor in a spatial context. The instructor shares the reference system
with the instructed system, as passenger of a wheelchair, easing the
interpretation of the spatial relations the instructions will comprise.
Since robotics maps typically consist of metric information while
instructions by humans mainly comprise qualitative relations, it is a major
goal of the project to combine quantitative and qualitative information for
purposes of robot navigation in the long term also for human-machine
communication about space.
|
