Do you want to program your own Super Mario bot, that makes use of the
latest AI techniques (or your favourite dirty hacks) to guide Mario
through challenging levels? Or do you want to create an automatic
procedural level generator, that creates new levels that provide
optimal entertainment for individual human players with particular
skills and preferences?
Are you ready to compete with researchers and programmers from around the world?
The 2010 Mario AI Championship will be held in conjunction with three
international conferences: EvoGames (part of EvoStar), IEEE World
Congress on Computational Intelligence and IEEE Conference on
Computational Intelligence and Games.
The competition is the extended successor to last year's Mario AI
competition, which attracted the attention of leading international
news media and from which the video of the winning entry has been
shown more than 600.000 times on YouTube.
This year, the competition has three tracks:
- Gameplay track, where the goal is to submit an agent that can make
Mario survive through a number of increasingly difficult levels. This
track is similar to last year's competition, but includes much more
difficult levels. Organized by Sergey Karakovskiy and Julian Togelius.
- Learning track, where the agent needs to learn how to play a
particular level optimally during a number of attempts. Organized by
Sergey Karakovskiy and Julian Togelius.
- Level generation track, where competitors submit level generators,
that generate fun levels according to player metrics and other
software-provided specifications. The levels will be tested "live"
on-site during the associated conferences. Organized by Noor Shaker,
Julian Togelius and Georgios Yannakakis.
The deadline for the EvoStar edition of the championship is April 1st;
deadlines for the WCCI and CIG editions are yet to be set, but will be
close to the relevant conferences (July and September).
Read more and download source code:
http://www.marioai.org
Please note that participation in the conference is not restricted to
academic researchers; professionals and students are also encouraged
to participate. Also note that the competition makes for excellent
class projects. We're looking forward to see your entry in the
competition!
--
Julian Togelius
Assistant Professor
IT University of Copenhagen
Rued Langgaards Vej 7, 2300 Copenhagen S, Denmark
mail: jul...@togelius.com, web: http://julian.togelius.com
mobile: +46-705-192088, office: +45-7218-5277
I'll probably try for something better this time around but on the
website it seems that only the Level Generation links are appearing.
One question, it seems if I write an awesome bought for the gameplay
track it should start out pretty good on the learning track. Though I
guess there are ways to make levels which don't follow the same
optimization patterns as regular game play. Like extra points for
killing a specific enemy, or being in a certain place when time runs
out, etc.
Also, I've determined the Java serialization api is much better then
the 3rd party library used last year. It's easy to use if you
implement the interface. ;)
Finding time for this with a full time job will be hard this time
around though. :P
Cool! Also, there will be (at least) two more deadlines Before WCCI and CIG.
> I'll probably try for something better this time around but on the
> website it seems that only the Level Generation links are appearing.
The other info (and code) is there - you just need to scroll down to
the bottom of the page... We're right now trying to sort out why
they're not appearing in the menu to the left.
> One question, it seems if I write an awesome bought for the gameplay
> track it should start out pretty good on the learning track. Though I
> guess there are ways to make levels which don't follow the same
> optimization patterns as regular game play. Like extra points for
> killing a specific enemy, or being in a certain place when time runs
> out, etc.
True, a good gameplay track submission should be the basis of a good
learning track submission...
> Also, I've determined the Java serialization api is much better then
> the 3rd party library used last year. It's easy to use if you
> implement the interface. ;)
WOX certainly has its share of problems, but standard Java
serialization has the cry serious problem that a serialized file won't
work if you change the source code at all - it's restricted to one
version of the code. This makes it unusable for our purposes, as far
as I understand.
> Finding time for this with a full time job will be hard this time
> around though. :P
Good luck!
I hope we're able to prevent Robin from resubmitting his bot and
winning two contests at once this time around.
> WOX certainly has its share of problems, but standard Java
> serialization has the cry serious problem that a serialized file won't
> work if you change the source code at all - it's restricted to one
> version of the code. This makes it unusable for our purposes, as far
> as I understand.
Yeah, I can see where the difficulties can arise (though, compatible
changes are allowed, for instance you can add a field). WOX wasn't
actually horrible in most cases, but it doesn't scale in space very
well, which didn't work well for my heavy handed approach. :)
Thanks for doing this again. Time to get to work!
On Mar 12, 10:36 am, Julian Togelius <julian.togel...@gmail.com>
wrote:
great! Thanks for spotting out!
Submission details fixed along with updates of all other pages
according to numerous private questions and mailing lists.
gl, hf,
Sergey
Fixed! thanks for pointing out;
> To unsubscribe from this group, send email to mariocompetition
> +unsubscribegooglegroups.com or reply to this email with the words
> "REMOVE ME" as the subject.
--
Kind Regards
Sergey
generaly speaking, in essense:
several levels would be choses and seeds published.
1000 trials to give opportunity for the agent to learn.
evaluation on the 1001 trial.
More details will be published on April, 3rd.
Sergey
On Mar 24, 2010, at 9:41 PM, xhunter70 wrote:
> Hi, What are the rules for the learning track, because the website
> doesn't help
>
I'm definitely interested in the Learning Track in this years
competition, but the website is still pretty light on details. While
your brief description below is good, I'm curious about a few other
things:
- What is the deadline?
- What exactly would be submitted? The final agent or the code to
create a learning agent?
- What defines a trial? A single run of the given level?
- For example, if I use a GA or GP approach and I've got a
population size of 500, would that mean I only have 2 evolutions?
TIA
On Mar 24, 4:47 pm, Sergey Karakovskiy <sergey.karakovs...@gmail.com>
wrote: