Machine Learning in Procedural Content Generation

In the context of video game development, procedural content generation (PCG) is often associated with asset creation, such as environments, or non-player characters. Video games like The Binding of Isaac and Enter the Gungeon utilise procedural generation to create gameplay variation, by connecting individual rooms created by a level designer to create a level. This project aims to explore the integration of Machine Learning (ML) in PCG to enhance quality and engagement in PCG outputs; these outputs being individual room levels of a top-down, 2D video game created in Unity.

Features:

• Procedural Content Generation Systems
  • Highly Expanded Tile System
  • A* Pathfinding PCG Generation
  • Developer Generation Tools
• Machine Learning Systems
  • Trained Reinforcement Learning Machine Learning Agents
  • Configurable Reward Structures
  • Machine Learning Training Environment
• Engagement Systems
  • Level Engagement Evaluation Environment
  • Level Saving System for Engaging Levels