[robocup-legged] Rule Changes and Poll

[Arnoud Visser]

At 09:48 AM 9/27/2006, Thomas Röfer wrote:
There were three Open Challenges about 
localization without the beacons (even harder: without the
goals as well), and I have the feeling that if teams have to localize
without them, they will be able to do so. However, I am not convinced that
passing will be ready in one year.

The Dutch Aibo Team demonstrated localization 
without beacons, goals or field lines.
This localization was based on learning the 
surroundings as a whole (without using natural landmarks).

At 09:01 PM 9/27/2006, Walter Nistico wrote:
>------------------------
>Not really.
>The flags positioned on the corners of the field are there just to reduce
>the ambiguity between a corner kick and a throw in when the ball goes out
>there nearby.
>They are not used by the football players to globally localize, they're
>barely visible, and if removed they don't affect the localization of the
>players at all (e.g. fog or heavy rain).
>Humans localize by looking at field lines, goals, teammates and opponents,
>and by observing natural landmarks like the audience etc, but to a much
>lesser degree.

I agree with this argument. The corner flags in 
human soccer do not play a role in localization.
The Dutch Aibo Team used only the observation 
above the top of the beacons. This means that 
there has to be enough visible above the beacons, 
and that the view on the audience is not shielded by a white wall.


At 09:29 PM 9/27/2006, Colin McMillen wrote:

>In real soccer (at least in professional games) there are significant
>advertisements present on a wall around the field.  In the past, we
>have had team logos, Sony logos, Aibo logos, etc. in the area nearby
>the fields, but always on the "outside" from the robot's point of view
>(on the outside of the walls back when we had walls, now on the sides
>of the raised field, etc.)
>
>It would be interesting to allow each team to make one logo or banner
>and place it in a (static) location around each field, in areas where
>the robots can see the logos and localize based on them if they so
>choose.  Since these "markers" are free-form and similar to what you
>might see in real soccer, they would increase the realism of the game
>while still providing some "natural" landmarks as additional visual
>cues to the robots.  (The crowd is much too unstructured -- and far
>away -- for the AIBO to do any meaningful localization directly on the
>audience, I think.)

In the Open Challenge we have demonstrated that 
localization on the audience is possible.

Logo and banners are a way to create natural 
landmarks. Yet, if they are located on field height,
they should be black and white. The approach of 
the Dutch Aibo Team is based on counting
color transitions, and would benefit from full 
color posters around the field (above beacon height).

Note that the Dutch Aibo Team didn't use this 
algorithm during the soccer games, because for a 
single learning spot only the orientation can be 
reliable estimated. Learning multiple spots can 
give an estimate for the current location on the 
field (+/- 20 centimeter), but not as accurate as 
our current algorithm (less 10 centimeter). Also 
the processing power between the different 
modules have to reconfigured before the algorithm can be used during the games.

Regards,

Arnoud Visser
Dutch Aibo Team