Jörg Denzinger's

AI methods for the development and set-up of computer games: automatic content generation

Automatic content generation, also known as procedural content generation, aims at using AI techniques (and others) to generate usually personalized game elements, like levels, stories or quests. Personalization to a particular human player is achieved by using a player's game playing history and any other information that a program can get from a human player.

Our work is around generating levels for platform games, like Super Mario or KGoldrunner. There are two general requirements we have to be after when creating a level:

  • the level must be solvable and
  • the level must be fun to play.
While what is fun is obviously very subjective (hence the need for personalizing the generation process), creating a solvable level, at first glance, should not be very difficult. For some games that is the case, but other games, where enemies can be rather clever, pose problems due to the resulting dynamic developments, that cannot be predicted just by looking at the static features of a level.

Our solution to this problem is using artificial players that try to solve a level (naturally combined with an analysis of static features to filter out as many level candidates as possible). Currently, we use only one play strategy for this artificial player, but customizations to human player types and even individual human players are possible.

We published our ideas at CIG 2011 and AIIDE 2012 (see our bibliography page).

to our ideas on automated character generation.

Last Change: 5/12/2013