CST-Godot: Bridging the Gap Between Game Engines and Cognitive Agents

Abstract

Cognitive architectures (CA) exist in the neighborhood of research in general AI. However, despite the high importance placed on classical AI techniques in developing complex videogame systems and the success the CA field has had in producing intelligent behavior of different kinds, it has gone largely unexplored in the context of videogame development. This paper presents a framework for implementing cognitive agents based on CAs in the Godot game engine. We employ the Cognitive Systems Toolkit (CST) for this purpose, providing opportunities for development in the academic sphere, primarily for creating experiments and visualizers that take advantage of Godot's physics and rendering libraries. In the context of videogame development, it can help expand upon classical AI techniques and enable more complex and interesting systems' implementation. As a concept proof, we show how a reinforcement learning-based agent can successfully learn how to behave in a game designed under this cognitive environment.