What is NERO?
NERO stands for NeuroEvolving Robotic Operatives. NERO is a game
being developed at the Digital Media Collaboratory
(DMC) at UT Austin that uses NEAT as its core AI technology.
NERO promises to change the way video games utilize AI.
Here is what makes NERO unique:
- AI agents evolve in real time while the game is being played
- Instead of playing against the AI, the player's role is to
train the AI for competition
- Since a special real-time version of NEAT is the core technology,
AI players have real neural networks that are continually growing
more complex as the game is played
- AI agents are not prepackaged or scripted; rather, their behavior
emerges in reaction to how the game is played
Where did NERO come from?
I first proposed the concept behind NERO at the
2nd Annual Game
Development
Workshop on
Artificial Intelligence, Interactivity, and Immersive
Environments, which was hosted by the Digital Media Collaboratory of the IC2 Institute at UT Austin. I
had been thinking about how to make evolution an entertaining part
of the video game experience. In particular, I had two nagging
questions in mind:
- If agents are allowed to evolve during a video game, what
prevents them from being wiped out before they achieve a suitable
and interesting level of play?
- How can we guarantee the game will be interesting when we
can't be sure what level of performance the AI will ultimately
reach?
After thinking hard about these questions, I came up with the concept
of NERO as the solution. I brought it up during a breakout session
during the Workshop. In the breakout session, a group of conference
attendees discussed and refined the idea, and I then presented the
idea to the Workshop at large. The DMC lab took interest and
decided to sponsor the idea as a project. More than 20 people
are currently involved.
What is going on in the screenshots?
These are really old screenshots from when NERO 1.0 was still in
early production.
The screenshots show a working development version of NERO in which
robots have been trained to pursue and attack an "enemy" (the word
"ENEMY" can be seen above its head). The green characters over
their heads in the first screenshot describe the brain complexity of the particular
individual. For example, "N: 21 L:54 S:75" means the robot's
neural networks includes 21 neurons, 54 connections, and is
in species #75 (species are part of
the NEAT
method of neuroevolution). The sliders are used to tell the
robots how we want them to behave, i.e. they adjust the fitness
function in real time.
Feel free to contact me for more info or
to comment.
