F.E.A.R.s AI was designed to provide a combat experience as exhilarating as multiplayer against a team of skilled humans. Enemies work in squads, laying suppression fire as allies advance. When confronted with a threat, soldiers shout orders to each other, and if necessary dive crashing through a window to safety. Soldiers advance in formation, split up into organized search parties, and flush out the Player with grenades.

The typical game AI techniques of the past could not meet the demands of the complexity desired for FEAR. Finite State Machines and A* are core AI systems in many games, and F.E.A.R. is no different. What sets F.E.A.R. apart is the fact that the Finite State Machine has only three states, and A* is used to plan actions as well as paths. Planning behavior in real-time allows characters in F.E.A.R. to adapt to the situation at hand. If the Player slams a door on a pursing enemy, the soldier can dynamically re-plan and decide to fire through the window, or flank through an alternate entrance. Empowering characters with real-time planning and problem solving abilities frees developers to focus on higher level squad behaviors, such as suppression fire, advancing cover, and search behaviors.

This talk describes techniques applied on F.E.A.R. that allow developers to use Goal Oriented Action Planning systems in real-time action games, and the benefits of doing so. It also describes how to make a planning system practical for real-time, and illustrates how autonomous planning characters support higher level squad behaviors. The session concludes with a discussion of current bleeding edge game-related AI research, and how planning can support communication between characters, or even between characters and the Player.