Research under Professor Steven Engels at the University of Toronto on deep-reinforcement-learning-driven procedural content generation and autonomous QA for generated levels.
This research project, conducted under Professor Steven Engels at the University of Toronto, explores how deep reinforcement learning can support game development through two parallel paths: generating content and autonomously evaluating it.
The first path plans to train a procedural content generation (PCG) agent through deep reinforcement learning. Rather than optimizing only for whether a level can be completed, the agent will target a specific dimension of game immersion and explore how that objective can guide the generation process.
In parallel, the project plans to develop autonomous QA methods for evaluating and validating the levels produced by the PCG agent. This path focuses on creating a repeatable way to inspect generated content and determine whether it satisfies the project's intended requirements.
Together, these paths investigate how deep reinforcement learning can contribute to both sides of procedural game development: producing levels designed around a player-experience goal and building autonomous systems capable of validating the resulting content.