Artificial intelligence takes on Ms. Pac-Man
A competition out of the University of Essex challenges gamers around the world to create AI that can beat Ms. Pac-Man.
For the last 30 years, Pac-Man's estranged wife, Ms. Pac-Man, has been consuming countless coins and hours from gamers at arcades across North America.
Philipp Rohlfshagen, David Robles, and Simon Lucas, all of the University of Essex, recently launched a competition called Ms. Pac-Man vs. Ghost Team for those wanting to rewrite the flow of the classic chomper. The contest, held last week in New Orleans at the Congress on Evolutionary Computation 2011, was an open call for savvy Java programmers to create artificial intelligence routines that either control where Ms. Pac-Man or the four ghosts should go during gameplay. If you can't beat them, hack them.
Programming a successful strategy for a game like Ms. Pac-Man in real time is very complicated (click here for the software). A more competitive version of Ms. Pac-Man was used in the exhibition; fruits were omitted and the speed of Ms Pac-Man and the ghosts were identical. Entrants to the competition weren't allowed to use screen captures to navigate. They were required to directly interface with the game engine instead.
According to the competition site, every second counts:
At each game tick the controllers (for Ms Pac-Man and for Ghost Team) are provided with the current state of the game. For the competition the game runs in real time with approximately 25 game ticks per second. Each controller will run in a separate thread and can take as long as it wants to respond but the game engine proceeds with the game after 40ms; if no action is supplied, the previous action is repeated. Otherwise, the game updates its state based on the most recent outputs (joystick directions) from each controller. Each game is limited to a certain number of game ticks to prevent stale-mates.
So what was the highest programmed score? AI Ms. Pac-Man programmed by Atif from University of Essex racked up 69,240 points against AI ghosts programmed by DarkRodry. However, DarkRodry must've not programmed a very good set of ghosts as the top four scores all are against him. James from the University of Nottingham had the highest average score for controlling Ms. Pac-Man with 34,278. The video below shows off what an AI Ms. Pac-Man game looks like.
The competition will return in late August for the 2011 IEEE Conference on Computational Intelligence and Games in Seoul, with more details coming to the Ms. Pac-Man vs. Ghost Team Competition site soon.