Connectionists: NERO 2.0 machine learning game (www.nerogame.org)

[Risto Miikkulainen]

We are pleased to announce the release of NERO 2.0 machine learning
game. In this game, the player trains teams of agents to perform
complex tasks in a simulated 3D environment. The agents are controlled
by neural networks that learn based on the rtNEAT neuroevolution
method.  The training is evaluated in autonomous battle mode against
other teams; the game also provides a territory-control mode for
interactive game play. The territory mode is new in 2.0; this release
also includes a new user interface and more extensive training tools.

NERO can be downloaded freely from http://nerogame.org for Linux, OS X,
and Windows platforms. It is intended to serve three purposes:

- It is an engaging game that demonstrates a new genre of video games
  where machine learning plays a central role. The site includes
  videos illustrating the gameplay and evolved behaviors, and the game
  includes a tutorial mode that makes it easy to get started.

- It is a "killer application" of rtNEAT, demonstrating how it can be
  used to learn complex behaviors in real time. For more details on
  rtNEAT and its application in NERO, see the paper at
  http://nn.cs.utexas.edu/keyword?stanley:ieeetec05. The rtNEAT (and
  NEAT) software is available at http://nn.cs.utexas.edu/soft-list.php.

- It is a prototype of a research platform that will allow developing
  and testing new machine learning methods in a complex video game
  environment, as well as a demonstration tools for various AI methods
  in general.

We would like to get your feedback especially on this last point. In
the near future, we will put together an open-source version of NERO
(v2.0 is based on the Torque game engine) and plan to extend it to
serve as a general research platform for the community.  How can the
NERO environment best support research in machine learning and
embedded artificial agents? How can it best serve as a demonstration
platform e.g. for AI courses? At this point, we invite you to try out
NERO 2.0 and give us feedback and suggestions on how to make OpenNERO
a useful such tool for the future.

-- Risto Miikkulainen, Ken Stanley, Igor Karpov,
   and the NERO development team