AW: [robocup-small] End of Community discussions

David M. Chelberg chelberg at endor.cs.ohiou.edu
Fri Sep 9 10:40:41 EDT 2005


In message <000001c5b547$86ae9120$796e2da0 at pcpool.mi.fuberlin.de>, "Raul Rojas"
 writes:
>> The time alloted for general discussion by the community is over. 
>
>The discussion was just starting to warm up...  Let me briefly go over some
>points. It is my personal feeling about all this:

I agree we just started our school year, so haven't had time to
discuss these ideas with our whole team, so let me state some of my
personal opinions.

>
>1) All questions related to speed of ball/robots/chip kicks etc. can be
>solved with automatic refereeing. The referee box gets the vision
>coordinates
>from any team through an Ethernet cable. The referee box can then compute
>the energy spent by a robot running, kicking, etc. We assign costs to
>running, kicking flat, kicking high, using some magic constants, and each
>robot gets an energy budget. The referee box keeps track of costs and
>signals
>infractions. A team with good vision is selected at each field to deliver
>the
>coordinates of the robots during a game.
>

I am not as convinced that auto refs will solve all these problems.
However I think that we should investigate this as a possibility.  I
also like the idea of going back to 45 degree walls, with perhaps a
more compliant ball to help with the arms race in kicking/chip kicking
technology. 

I think we might want to have an adjunct auto refereeing competition
for the next robocup.  You could turn on your auto ref when you
weren't playing (you can use your own cameras).  A year or two
developing auto refs in a separate but related competition might spur
their development and help us find bugs.  I think without this amount
of testing the autoref idea will probably be a disaster.

>2) The small-size is not the entry league to RoboCup. That is just not
>happening.
>Smalls-size is more expensive than the AIBOs. The AIBO league is the entry
>league,
>and there are lots of code on the net for the standard robot they use. There
>are
>more and more AIBO teams because of this.

Agreed.  It is difficult to start in small-sized league right now
given the high level of technical success of top teams in speed,
agility and kicking.

>
>3) The small-size league must not exist forever, certainly not until 2050.
>At
>some point it will disappear, once it has run out of steam. I don't expect
>the
>small-size league to exist many more years.

I think there are plenty of potential research areas in small-sized
league.  I don't think we've exhausted AI challenges, and if we go to
local vision keeping the speed and agility we currently have will be a
great challenge.

>
>4) I think that the research challenges have been exhausted with the current
>
>format. The only two new challenges that could make small-size more exciting
>again would be automatic refereeing, with minimal human intervention, and a 
>larger field.

A larger field for greater team play and passing would be nice in the
longer term, and local vision would also make the game more
interesting.  We could keep global vision for auto ref purposes as we
go to local vision.  Also, having our robots fully self-contained has
additional distinct challenges.  This would be possible in the short
term if we either had a vision server for all teams, or went to local
vision.

The combination of many (7 perhaps moving to 11 (see next comment))
small robots, large field, full autonomy and auto ref. would keep us
distinct from other leagues.

>
>We play now on a field which is almost a factor four larger than in 1999. In
>2000/01
>there was much criticism that a larger field could not be used. We did and
>we
>progressed.
>
>We could go again for a factor four of increase, to play in a mid-size
>field. That would imply
>that only four cameras are used and all teams get the same video signal, or
>that
>a standard vision system is used by all teams, so that a single set of four
>cameras
>is enough for all teams. I know it is difficult at first, but rewarding at
>the
>end. Global vision could be used additionally by any interested team.
>
>5) 11 on 11 on a mid-size field would again make the field very crowded. My
>back
>of the envelope calculation is that 7 on 7 is ideal, and it keeps cost down.
>Automatic
>refereeing could also restrict how crowded the field is (you would not be
>allowed,
>for example, to move all robots at once to a point on the field, etc).
>

See above.

>Please note that the exodus has begun. Good teams are leaving because the
>research
>challenge has reached the point of "diminishing returns". If the small-size
>league
>is to survive for another few years (and no more than that), then the bar
>has to
>be raised again.

I agree we have to find ways to keep the research interesting for both
the top teams but also for teams still striving to reach that level.
One way to achieve this is by having multiple research goals that we
can pursue at the same time -- e.g. AI team play, opponent modeling,
local vision, auto refs., etc.

 -- Prof. David Chelberg (chelberg at ohiou.edu)
    Team RoboCats (Ohio University)



More information about the robocup-small mailing list