General AI for tabletop games
TAG: Tabletop Games Framework
An open-source Java framework for implementing modern tabletop games, connecting them to AI agents and studying their mechanics and gameplay.
Why this project is interesting
Modern board, card and dice games combine changing rules, diverse player roles, hidden information and many interacting components. TAG provides a shared environment for studying these challenges with general AI methods.
The problem
Much game-AI research has focused on classical games, while modern tabletop games introduce varied mechanics and information structures. Researchers also need consistent game and agent interfaces to compare methods and inspect game properties.
Approach
TAG is a Java benchmark with a common API for AI agents, reusable components for implementing games, JSON-based game data import and logging for analysis of measures such as action space, branching factor and hidden information. The 2020 design report describes seven implemented games at that time, but the framework now includes over 60 titles.
My contribution
I am one of the founders and leaders of the TAG project, and I have designed and maintained the core parts of the framework including game structure, API, AI agent interaction, game running, user interfaces and metrics recording. I have implemented several AI agents (e.g. RHEA) and games (e.g. Terraforming Mars, Pandemic), and I have coordinated, reviewed and revised contributions from other developers and researchers. I have led courses and student projects using the TAG framework as a basis for research and development, and I have presented the framework at conferences and workshops such as the IEEE Conference on Games.
Outcomes and evidence
- A public Java framework and source repository support repeatable AI research in modern tabletop games.
- The framework paper introduced sample AI players and outlined research directions; it also notes substantial room for improvement.
- Related studies have used TAG for individual games and AI tasks; add specific examples and verified adoption evidence as appropriate.