A Survey on the Development of Human-Centered Artificial Agents for Fighting Games

Artificial intelligence methods can be used to develop artificial agents for competitive video games. These agents have the potential to be learning tools for players, as well as provide automated insights into how balanced the game is. This motivates the existence of agents whose skill is adjusted to the player. However, this also assumes the agents are good substitutes for human players. Developing “human-centered” agents, focused on the gameplay experience of human players, is a difficult problem. In the genre of fighting games, they need to account for different characteristics such as human-like adaptation and reaction time, which is an important factor in these games' design. We survey the different dimensions by which this problem is addressed, both in commercial games and the literature. These dimensions include the performance of the agents, their human likeness and the diversity of their behaviors. We thus present the state of the art in the creation of human-centered agents, along with future directions in which this problem can be explored further.