I am of the view that the goal of this competition is to develop both
agents and simulators in a way that can be applied in the real world. In
the real world we have certain knowledge about how fires spread, which
roads are important etc, but we certainly do not know precisely how a
particular building is going to burn, or where everyone in the city is at
each moment in time.
I think that, in general, "good" simulators will have some elements of
randomness in them because the real world has randomness (or things that
appear random). Therefore it is possible to allow agent developers access
to the source code of the simulators without letting them write hard-coded
solutions. The use of randomised maps will also alleviate this problem.
>From a fairness point of view we really have to allow everyone free access
to simulator source code - as Mohsen points out: "We are also agent
developers...we are a little different [from] other teams" - to allow one
team to view the source code but not any other team makes the competition
unfair. In the worst case it allows simulator developers to build "cheats"
into the simulator (for example, I can easily write a fire simulator that
has a cheat: extinguishing with 127 units of water puts out every fire in
the city).
I therefore don't have a problem with allowing agent developers access to
all the internal workings of the simulators. The open source policy we
already have will accomplish this.
That's my two cents.
Cameron Skinner
The Black Sheep
Department of Computer Science
The University of Auckland
This archive was generated by hypermail 2.1.3 : Wed 21 Apr 2004 - 22:31:46 GMT