Monday, January 19, 2009


We had problems to find a place where the compass worked stable. This means there should be a homogeneous and static magnetic field, not necessarily depending on the magnetic field of the earth. The map needs to have the same direction as the compass.

At first the robot is doing a calibration of the compass, then connection to the pc. The map is aligned to compass zero direction. After this phase the robot is placed randomly on the map. This time their are no zero rounds performed to make the algorithm find the robot first and then navigate. The activities are performed both at the same time. As it is shown in the posted video the samples need several rounds of robots driving to build a cluster at the position. There is a second cluster with is close to the one at the real position. After some more steps the cluster surrounds the real robot position.

The MCL2D algorithm worked well in an interactive environment and was able to localize the robot after some rounds.

The results in the second map are not as good as the one in the other. In the experiments did on the second map were always more then one cluster, the robot was lost in a few tests. Sometimes it worked well. It is to say that the second map has to much ambiguities to localize proper. We tried to spread some samples (about 5-20%) in the map to find the robot again when it was lost while the localization process. After this change the algorithm didn't work well anymore. So the change was skipped.

No comments: