Reproduce “Automated Game Design via Conceptual Expansion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Matthew Guzdial, Mark Riedl
Researchers at Georgia Tech's School of Interactive Computing developed an automated game design system that learns game mechanics and level design principles directly from gameplay video footage. This system then generates entirely new games by recombining these learned representations using 'conceptual expansion,' achieving lower combined error than baseline methods when recreating classic NES platformers from partial specifications.
Key insights
- It's possible to learn both structural level design and functional game rules directly from raw gameplay video footage without requiring pre-authored knowledge.
- A unified 'game graph' representation can effectively integrate spatial relationships, sprite properties, and rule-based interactions within a single framework.
- The process of inferring level design from given game rules is generally more challenging than inferring rules from given level design, aligning with intuition about how constraints operate in game design.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.