Abstract
This thesis addresses path planning in changeable environments. In contrast to traditional path planning that deals with static
environments, in changeable environments objects are allowed to change
their configurations over time. In many cases, path planning algorithms
must facilitate quick answers to queries in order to be useful. For
example, an opponent in a
... read more
training simulation needs to respond to the
actions of the user without any significant delay. To achieve such
performance, path planning methods usually use a preprocessing phase in
which the environment is explored. As much computation time as possible
is moved to this preprocessing phase such that at query time only little
time is needed to solve an actual path planning query. This approach has
led to many successful methods that are applicable to a broad range of
problems.
Because of the nature of preprocessing, existing methods have difficulty
to cope with unanticipated changes that occur in the environment in a
later stage. Often existing solutions are computationally expensive and
may fail if no local solution exists.
This thesis presents novel results for path planning in changeable
environments. It is divided in three parts. The first part deals with
the class of problems in which obstacles can change their configuration
between the time the roadmap was created and the query. Examples of such
obstacles are doors, chairs and boxes. We provide an algorithm that is
able to deal with such changes in the environment while keeping the
planning process efficient.
The second part deals with environments in which robots have the ability
to manipulate obstacles that block their path. Imagine, for example a
simulation in which a firefighter commander is trained. The commander
gives his (virtual) firefighters higher level commands (e.g. "walk
around the building and enter it at the back"). For a realistic
training, the firefighters should be able to move away obstacles that
block their paths in order to, for example, clear the door. The
algorithms in this part describe a novel way to deal with this type of
problems by imitating human behavior.
Finally, the third part deals with the problem of a robot pushing a disk
in a polygonal environment. Pushing an object by a robot is often easier
or more applicable than pulling since it does not involve grasping the
object. A robot arm can push an object using a single finger while
pulling involves more complicated behavior. Unfortunately in addition to
the usual sensor errors, pushing is also sensitive to another type of
uncertainty; if the object's center of mass is not exactly known then
pushing an object leads to erratic behavior leading to unstable pushes.
We provide solutions that are robust against sensor errors and therefore
are more suited to be used in practical problems.
show less