Dear Technical Committee,
we (ResQ Freiburg) appreciate your efforts to advance RoboCupResuce -
however, Sebastien is right in all his claims, and many more can be made:
Day 2:
> change of the sensing property. First of all, it goes against the rules
> that were adopted for the competition. Changing the rules less than
> three weeks before the competition is really not good.
That is the most important point. What you propose is a FUNDAMENTAL
change of the competition only a few weeks before its start. The teams
will only be able to come up with some weak ad-hoc solution and will not
be judged for what they did during the last year.
Especially when no messages at all are allowed, we are facing a
completely different problem. I cannot imagine how one should adequately
prepare for this "challenge" in the remainig time.
> If there is still gone a be a sensing competition, we need more specific
> information to adapt our program:
Absolutely. In addition to Sebastien's questions about communication,
especially what you mean by "reducing visibility" is not clear: is it
only the visibility RANGE that is going to be reduced or do you speak of
RELIABILITY of sensing (e.g. if an agent sees an empty house there might
still be a probability that there is a Civilian inside)? The latter
variant again would make all reasoning efforts invalid (resp. would
require probabilistic reasoning or the like).
> And now, for the learning part of the competition. I think it is
> impossible to do useful learning in only three simulations. The only
> thing possible in such a short time is simply to record everything you
> can so the agents won't have to search for the fires or the civilians.
> This is not interesting learning since there is no generalization, it is
> simply "hard coded" learning for a specific situation. The information
> learned would not be interesting for other situations.
Exactly. And the ranking scheme you propose seems to enforce the
following strategy: simply do nothing in the first two rounds (or just
explore). Then in the last round try to score points in order to have
maximal difference between the 1st and 3rd run. Well... ;-)
Maybe I just didn't understand what "the same random map" is: same city?
same *polydata.dat? same gisini.txt? same random seed? all of these?
To sum up: 1. The new challenges are interesting and important. 2.
Currently, lots of things are unclear about how these challenges are
going to look like. 3. Most teams must leave for Lisboa in about two
weeks. Conclusion: Why not postpone this to RescueMiddleEarth? Or next
year?
We very much appreciate most of the ideas of the committe, like random
maps and gisini files completely unknown to the competitors, that is,
things that prevent hard-coding and too much problem-specific tuning.
However, we believe that changes as fundamental as the ones you propose
should have been discussed half a year ago (like stated in the rules
decision timetable).
Michael (for ResQ Freiburg)
-- Michael Brenner Phone: (+49) (761) 203-8226 Institute of Computer Science Fax: (+49) (761) 203-8222 Albert-Ludwigs-University mailto:brenner@informatik.uni-freiburg.de Am Flughafen 17, Geb. 52 http://www.informatik.uni-freiburg.de/~brenner D-79110 Freiburg, Germany
This archive was generated by hypermail 2.1.3 : Thu 10 Jun 2004 - 16:21:06 GMT