GAR is both a multiplayer online video game and an experiment in
automatic content generation driven by player preferences. Unique game
content, namely spaceship weapon systems, automatically evolves based
on player behavior through a specialized algorithm called cgNEAT
(content-generating NeuroEvolution of Augmenting Topologies). In
particular, new variants of weapons that players like are continually
created by the game itself. In this way, the game never stays the
same. GAR is the first video game to demonstrate that critical content
can evolve as the game is played to satisfy its players. For an idea
of what the game is like, think “Space Diablo”, with real-time combat,
cool particle system weapons, and RPG-style leveling and skill trees.
We wanted to demonstrate that unique AI technology can positively
impact how video games are made. We chose automatic content
generation as the main focus because we believe that content is a
major bottleneck in the industry today. While most effort in AI in
games has traditionally centered on NPC behavior, content generation
may prove a most compelling alternative application for AI.
Galactic Arms Race generates its own content (in particular, weapons)
through an evolutionary algorithm in real time while the game is
played. Players can constantly discover unique weapons spawned in the
online galaxy, evolved through their past weapon preferences. To our
knowledge, GAR is the first demonstration that this principle can work
in a multiplayer video game.
The game took about 12 months to complete the first release but the
genesis of the cgNEAT AI and particle system evolution technologies
stretches back as far as 10 years to the invention of the original
NEAT algorithm. We are currently working on GAR version 1.2.
Before the game was even started, we spent about a year experimenting
with evolving particle systems (for genetic art programs) to learn how
they can be generated most effectively by an evolutionary algorithm.
Several interesting facts:
-GAR won the Best Paper Award at 2009 IEEE Symposium on Computational
Intelligence in Games (CIG09) and was Editor’s Pick for Best AI in an
Independent Game at AIGameDev.com.
–GAR has been covered in a variety of online media sources including
Slashdot and Le Monde.
-Finally, Ken Stanley, who directed the development of GAR along with
Erin Hastings, was also a leader on the NERO video game, another experimental game based on evolutionary
In evolutionary computation, which is the subfield of AI that studies
evolutionary algorithms, the term “arms race” has a special meaning.
At the same time, collecting weapons in outer space is also literally
an “arms race.” So the title of the game actually manages to describe
two facets of the game at the same time.
GAR co-leads are Erin Hastings and Ken Stanley.
Erin – Port Saint Lucie, FL
Ken – Newton, MA
Erin – Undergrad: University of Florida, Grad: University of Central Florida
Ken – Undergrad: University of Pennsylvania., Grad: University of
Texas at Austin
Erin – Graphics and gaming related research for graduate work. A lot
of mods and maps for Unreal Tournament. GAR is my first original game.
Ken – Long-time researcher in AI in games. First major production was
NERO in 2005.
The GAR team met in the Evolutionary Complexity Research Group at the
University of Central Florida. Ken Stanley is the research group
leader and the rest of the team is current or former UCF students.
Erin – World of Warcraft, Call of Duty Modern Warfare , Rock Band 2, Castle Age
Ken – New Super Mario Bros. Wii
Erin – Most hours spent probably in Everquest or WoW.
Ken – Galactic Empire, an online multiplayer BBS game from around 1990.
Erin – Whack-A-Mole
Ken – Any game that simply exploits old ideas to make more money.
Erin – Super NES
Ken – ColecoVision
Erin – PC
Ken – Wii
Erin – AI Research.
Ken – AI Research.