Lehigh UniversityStephen Lee-Urban
Imagine designing challenging and flexible AI team behavior for a first-person shooter (FPS) game. Among the decisions that must be made are whether to use static or dynamic strategies, whether or not these strategies should be represented symbolically, whether the AI should be able to learn, and, finally, the extent to which the actions that are taken by individual team members will be controlled by the team strategies.
This article presents an approach, called RETALIATE, which addresses these questions using an online reinforcement learning (RL) algorithm. In ...