Architecture for Growth of AI-Driven Characters Along Play

Abstract

This paper proposes an architecture with a more flexible character growth along play, making the characters not only able to grow differently based on how the player interact with them, but also act independently, and follow its needs, its goals and state of mind based on what he is evolving toward. This proposed architecture contains a decision making system based on a utility-based goal manager and planner. It allows the AI-driven character flexibility in choosing a goal to achieve based on a score of interest computed for each of them at any time. Then, the planner is used as a tool to find a series of action to satisfy the goal's objectives, allowing separation between goals and actions, and allowing the state of mind of the character, its knowledge and what is happening around it to influence the plan of action. Thanks to a learning component, the character is also able to learn how to use objects from experiences, but also inherit this learned knowledge to new objects it encounters based on their similarity. By combining all these technologies into one architecture, we think we have pushed character growth along play further.