Co-creating Game Content using an Adaptive Model of User Taste
Antonios Liapis, Georgios N. Yannakakis, and Julian Togelius
Center for Computer Games Research
IT University of Copenhagen
Rued Langgaards Vej 7, 2300 Copenhagen, Denmark
{anli, yannakakis, juto}@itu.dk
Mixed-initiative procedural content generation can augment
and assist human creativity by allowing the algorithm
to take care of the mechanisable parts of content creation,
such as consistency and playability checking. But it can also
enhance human creativity by suggesting new directions and
structures, which the designer can choose to adopt or not.
The proposed framework generates spaceship hulls and
their weapon and thruster topologies in order to match
a user’s visual taste as well as conform to a number of
constraints aimed for playability and game balance. The
2D shapes representing the spaceship hulls are encoded
as pattern-producing networks (CPPNs) and evolved in
two populations using the feasible-infeasible 2-population
approach (FI-2pop). One population contains spaceships
which fail ad-hoc constraints pertaining to rendering,
physics simulation and game balance, and individuals in
this population are optimized towards minimizing their distance
to feasibility. The second population contains feasible
spaceships, which are optimized according to ten fitness
dimensions pertaining to common attributes of visual taste
such as symmetry, weight distribution, simplicity and size.
These fitness dimensions are aggregated into a weighted
sum which is used as the feasible population’s fitness function
— the weights in this quality approximation are adjusted
according to a user’s selection among a set of presented
spaceships. This adaptive aesthetic model aims to
enhance the visual patterns behind the user’s selection and
minimize visual patterns of unselected content, thus generating
a completely new set of spaceships which more accurately
match the user’s tastes. A small number of user
selections allows the system to recognize their preference,
minimizing user fatigue.
The proposed two-step adaptation system, where (1) the
user implicitly adjusts their preference model through content
selection and (2) the preference model affects the patterns
of generated content, should demonstrate the potential
of a flexible tool both for personalizing game content to an
end-user’s visual taste but also for inspiring a designer’s creative
task with content guaranteed to be playable, novel and
yet conforming to the intended visual style.
Related Work
A. Liapis, G. N. Yannakakis, and J. Togelius, “Adapting Models of
Visual Aesthetics for Personalized Content Creation,” IEEE Transactions
on Computational Intelligence and AI in Games, Special
Issue on Computational Aesthetics in Games, 2012, (to appear).
A. Liapis, G. N. Yannakakis, and J. Togelius, “Optimizing Visual
Properties of Game Content Through Neuroevolution,” in Artificial
Intelligence for Interactive Digital Entertainment Conference,
2011.
A. Liapis, G. N. Yannakakis, and J. Togelius, “Neuroevolutionary
Constrained Optimization for Content Creation,” in Computational
Intelligence and Games (CIG), 2011 IEEE Conference on,
2011, pp. 71–78.
Figure 1: The fitness dimensions used to evaluate spaceships’ visual
properties and sample spaceships optimized for each fitness
dimension. Weapons are displayed in green and thrusters in red.
Figure 2: The graphic user interface for spaceship selection.
International Conference on Computational Creativity 2012 230