CS Doctoral students study opponent modeling in poker
Picture yourself driving in rush hour. Not only do you have to mind the rules of the road, but you have to consider the other drivers as well. You might switch lanes to avoid the speedster that is coming up behind you. Or, perhaps you’ll pass the overly cautious new driver ahead.
Wouldn’t it be nice to have an autonomous vehicle that could drive for us? Well, in order for this to happen, Robocar would have to assess situations on the road and make driving decisions quickly. Creating an agent that is able to observe others and act accordingly would be an integral part of this system.
As this is an exercise in reasoning, we can once again use games as our practice field. Nolan Bard and Mike Johanson are studying and reacting to their opponents in a Heads-Up (two-player) Texas Hold’em poker game. They realized that different styles of play (aggressive versus passive, for example) work better against different challengers.
“The idea is to take advantage of the mistakes your opponent is making,” says Mike. “In poker, the value is more apparent, however, these ideas could be applied to a broader spectrum as well.”
Nolan and Mike are working on two different components of opponent modeling. Mike is developing techniques to create static players that are both reliable and exploitive, while Nolan is working on methods to observe opponents and dynamically decide which static player to use during play.
Together, they are hoping to take Polaris (the U of A’s current poker playing champ), and AI in opponent modeling, to the next level.
Photos, 2010; article 2011.