Ludii General Game System for Modeling, Analyzing, and Designing Board Games

Cameron Browne, Eric Piette, Matthew Stephenson, Dennis J. N. J. SoemersEncyclopedia 2024 · DOI · OpenAlex

Introduction

Ludii is a "general game system" designed to model, analyze, and design thousands of different board games using a unique, high-level logic based on ludemes—the fundamental building blocks of play. By providing a unified language and a high-performance engine, it serves as a powerful bridge between historical game research, modern game design, and cutting-edge AI development.

In the field of computational ludology, most systems are specialized; an engine built for Chess is typically useless for Backgammon. While "General Game Playing" (GGP) systems exist, they often rely on low-level, verbose logic that is difficult for humans to write and inefficient for computers to process. Ludii addresses this by creating a system that represents a vast array of games—from ancient historical reconstructions to modern puzzles—using a succinct, human-readable format. It successfully balances generality (the ability to play any game) with efficiency (the speed required for AI to simulate millions of moves).

The Ludemic Model and Architecture

The core of Ludii is the ludemic model. Instead of describing a game through raw logic, it uses "ludemes"—pre-defined game concepts like "board," "piece," "slide," or "win condition." This high-level abstraction allows complex games to be described in a fraction of the code required by older GGP languages. For example, a full definition of Tic-Tac-Toe or the Game of the Amazons fits on a single page.

The system's modular architecture separates game logic from user interaction and decision-making, as illustrated in the following diagram:

The modular architecture of the Ludii system

Key technical components include:

  • Ludii Game Description Language (L-GDL): A LISP-like language where games are defined as trees of ludemes.
  • Class Grammar Approach: The grammar is automatically generated from the underlying Java code, ensuring a 1:1 correspondence between the description language and the engine's internal logic.
  • Modular Design: The system separates the game logic (Core) from the user interface (Player) and the decision-making (AI). This allows researchers to swap out AI agents or visualization styles without modifying the underlying game rules.

The Ludii Game Library and Database

The primary strength of Ludii lies in its extensive library, which contained over 1,000 predefined games at the time of publication. This includes traditional tabletop games, dice games, mathematical puzzles, and even simple video games.

A sample of the diverse games modeled within Ludii

Beyond the library, the paper introduces the Ludii Game Database (LGD). This database stores not just the rules, but also "evidence"—archaeological and historical data used to reconstruct ancient games where the original rules may have been lost. This makes Ludii an invaluable tool for historians and archaeologists seeking to preserve cultural heritage through digital simulation.

Technical Innovations and Performance

Ludii introduces several key insights for efficient general game playing:

  • Optimized State Representation: Ludii employs different internal representations (Flat, Graph, or Stack-based) depending on the game's needs to minimize memory usage and maximize reasoning speed.
  • The "Forward Model" API: Ludii provides a standardized API that gives AI agents a "forward model"—the ability to simulate future states. This makes it a plug-and-play platform for testing new reinforcement learning and search algorithms.
  • Broad Support and Performance: The system supports 1 to 16 players, stochastic elements, hidden information, and both adversarial and cooperative modes. It is efficient enough to support state-of-the-art AI techniques like Monte Carlo Tree Search (MCTS) across its entire library.

Conclusion and Future Directions

While Ludii is incredibly versatile, it is primarily optimized for discrete, board-based games. It is not intended to replace engines like Unity or Unreal for real-time, continuous physics-based play. Future work involves expanding the library further and deepening integration with "deep" AI frameworks. The authors highlight Polygames, a framework for training deep learning agents, as a key direction for achieving superhuman performance across the diverse range of games Ludii can represent. By lowering the barrier to entry for game modeling and AI testing, Ludii stands as a foundational tool for the future of digital ludology.

Summary updated

BibTeX
@inproceedings{browne2024ludii,
  author = {Cameron Browne and Eric Piette and Matthew Stephenson and Dennis J. N. J. Soemers},
  title = {{Ludii General Game System for Modeling, Analyzing, and Designing Board Games}},
  booktitle = {Encyclopedia},
  year = {2024},
  doi = {10.1007/978-3-031-23161-2\_486},
}
Figure