acm-header
Sign In

Communications of the ACM

ACM TechNews

Knowledge Transfer: Computers Teach Each Other Pac-Man


View as: Print Mobile App Share:
Pac-Man

Researchers at Washington State University have developed a method that enabled a virtual robotic agent to teach a student virtual robot how to play Pac-Man.

Credit: Midway Games

A computer can give advice and teach skills to another computer similar to the way a real teacher interacts with a student, using a method developed by researchers at Washington State University.

A student virtual robot learned how to play Pac-Man and a version of the StarCraft video game, and surpassed the performance of the teacher robotic agent. The team programmed the teacher agent to focus on action advice, or telling the student agent when to act. The trick is in knowing when to give advice, considering the student agent will get annoyed if the teacher agent is always telling it what to do.

"We designed algorithms for advice giving, and we are trying to figure out when our advice makes the biggest difference," says WSU professor Matthew E. Taylor. He wants to develop a curriculum for the agents that starts with simple work and builds to more complex actions, as well as to develop a better way for robots to teach people.

The research was partly funded by the U.S. National Science Foundation, which also gave Taylor a grant to use ideas from dog training to train robotic agents.

From WSU News
View Full Article

 

Abstracts Copyright © 2014 Information Inc., Bethesda, Maryland, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account