This book presents some of the most relevant results from academia in the area of Artificial Intelligence for games. It emphasizes well theoretically supported work supported by developed prototypes, which should lead into integration of academic AI techniques into current electronic entertainment games. This book elaborates on the main results produced in Academia within the last 10 years regarding all aspects of Artificial Intelligence for games, including pathfinding, decision making, and learning. A general theme of the book is the coverage of techniques for facilitating the construction of flexible not pre-scripted AI for agents in games. Regarding pathfinding, the book includes new techniques for implementing real-time search methods that improve the results obtained through A*, as well as techniques for learning pathfinding behavior by observing actual players. Regarding decision making, this book describes new techniques for authoring tools that facilitate the construction by game designers (typically non-programmers) of behavior controlling software, by reusing patterns or actual cases of past behavior. Additionally, the book will cover a number of approaches proposed for extending the essentially pre-scripted nature of current commercial videogames AI into a more interactive form of narrative, where the story emerges from the interaction with the player. Some of those approaches rely on a layered architecture for the character AI, including beliefs, intentions and emotions, taking ideas from research on agent systems. This book also includes chapters on techniques for automatically or semi-automatically learning complex behavior from recorded traces of human or automatic players using different combinations of reinforcement learning, case-based reasoning, neural networks and genetic algorithms.
