This page will show some further examples of FootSLAM maps.
Our new building
In December 2011 we moved into our new building back on the main DLR campus at Oberpfaffenhofen. The building is different to our previous one in that it has no prominent loops and is quite long. We'll be adding more examples shortly, but here is the result of our longest data set (~30 minutes walk, 50.000 particles, 16 mins. processing time on a standard PC) and the odometry from which the map was reconstructed:
Another example, with a shorter walk and its corresponding raw human odometry is this one:
The figures with the human raw odometry illustrates how hard FootSLAM had to work, so to speak.
Walks at Massachusetts Institute of Technology (MIT) Stata Center
In March 2010, four walks were taken in the building designed by Frank Ghery to be processed using the "Turbo" FeetSLAM algorithm. 90.000 particles were used.
Walk in Vienna
Here is the FootSLAM map of a building in Vienna that has a rather interesting layout. Note that there are no major loops that join the three main arms. The map has been obtained from a single walk and processed with 50.000 particles.
Walk during the Munich Satellite Navigation Summit in the Residenz building
The Munich Satellite Navigation Summit took place in March 2011. We show here one of the walks around the premises in the Residenz building and a partial layout of the original plan (another example of incomplete maps ;-) ). In red we have marked the location of the toilets, which are not shown in the reference map available to us. Note that the angle of the long arm that points up was correctly learnt - it's actually not 90 degrees to the rest of the building but leans slighly towards the left. FootSLAM does not assume right angled building layouts.
Our old building
Our old offices were located outside the main campus of DLR and provided the perfect environment for closing loops. Five different walks taken between 2009 and 2011 were processed using the "Turbo" FeetSLAM algorithm with 90.000 particles. The total map obtained is shown in the figure. We have also depicted the walls as shown in the original plans and the existing furniture layout at the time of the walks.
With these data sets, FootSLAM was able to demonstrate its abilities to detect the right location of walls and other obstacles. In red, one original wall location has been depicted. The wall was actually moved and the plans were not updated. The arrow points to its actual location which we only discovered after FootSLAM made, what we though to be, repeated "mistakes" in that area.