Putting the magic in the machine since 1980.

Thursday, October 9, 2014

Automatic Generation of Agent Behaviors

This past Summer at AAMAS my PhD student Bridgette Parsons and I published a paper with an algorithm that takes raw observational data and builds an agent behavioral models. Basically, we have lots of data of what nurses do in a hospital (with timestamps) and used it to build an MDP-like model of the "average" nurse. Its not really an MDP because we have loops, and its not really "average" in the mathematical sense. You can get the paper.

Bridgette is now building a first-person shooter game which she will use to gather data on gamers' play. We will then use this data to build a player model. That is, we plan to automatically build NPCs that behave like real players.