We are using simulation as a play- and testing-ground for navigation strategies, hoping to get ideas we just would not get by only working with real robots. One of these ideas is discussed in more detail below.
|
There will always be one disadvantage in using simulators: unrealistic enviroment.
But on the other hand - working with real robots - one cannot simply abstract
from any sub-problem (e.g. object identification, obstacle avoidance), which
easily results in diversion from the key topic: navigation. We are using simulation as a play- and testing-ground for navigation strategies, hoping to get ideas we just would not get by only working with real robots. One of these ideas is discussed in more detail below. |
 | As mentioned above, unrealistic enviroment is the main problem of any simulation. It is not our goal to push forward the borderline of what can be simulated, moving only slightly towards reality. Instead, implementing the strategies for evaluation and veryfivation on robots is our aproach. |
 |
Offices are two-dimensional worlds, only containing one-dimensional
objects: walls. The unequivocal identification of walls is assumed
to be a solved problem, which is quite strong restriction. Though, tests have
shown that working with incorrect identification is also possible. Furthermore,
the enviroment may not be cyclic. Our virtual robot is equiped with a camera capable of recording a 360 degrees panorama view. As the world is two-dimensional, a scan results in a line of pixels. Camera recordings of a moving robot result in movies (as can be seen below), where the data is presented in each row from -360 to +360 degrees. |
| The task: | The algorithm: |
| Find the way back to a specific wall | loop select two visible objects around goal object turn between selected objects advance until goal object is visible |