Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dominion: A New Frontier for AI Research

Danny Halawi, Aron Sarmasi, Siena Saltzen, Joshua McCoy

TL;DR

The paper presents Dominion as a compelling reinforcement learning benchmark due to per-game card-set randomness and a robust online dataset, accompanied by a practical RL baseline. It introduces a large Dominion Online Dataset with over $2{,}000{,}000$ games and a Rainbow DQN-based bot trained via self-play, showing competitive performance against strong heuristic bots like Provincial. The work formalizes benchmark-friendly structures (buy menus) and evaluates competitive coevolution against evolved strategies, providing a clear pathway for future RL research in deck-building and combinatorial games. By combining data resources, a readable baseline, and analyses of existing bots, the study establishes Dominion as a scalable testbed with practical implications for designing and evaluating generalizable RL agents.

Abstract

In recent years, machine learning approaches have made dramatic advances, reaching superhuman performance in Go, Atari, and poker variants. These games, and others before them, have served not only as a testbed but have also helped to push the boundaries of AI research. Continuing this tradition, we examine the tabletop game Dominion and discuss the properties that make it well-suited to serve as a benchmark for the next generation of reinforcement learning (RL) algorithms. We also present the Dominion Online Dataset, a collection of over 2,000,000 games of Dominion played by experienced players on the Dominion Online webserver. Finally, we introduce an RL baseline bot that uses existing techniques to beat common heuristic-based bots, and shows competitive performance against the previously strongest bot, Provincial.

Dominion: A New Frontier for AI Research

TL;DR

The paper presents Dominion as a compelling reinforcement learning benchmark due to per-game card-set randomness and a robust online dataset, accompanied by a practical RL baseline. It introduces a large Dominion Online Dataset with over games and a Rainbow DQN-based bot trained via self-play, showing competitive performance against strong heuristic bots like Provincial. The work formalizes benchmark-friendly structures (buy menus) and evaluates competitive coevolution against evolved strategies, providing a clear pathway for future RL research in deck-building and combinatorial games. By combining data resources, a readable baseline, and analyses of existing bots, the study establishes Dominion as a scalable testbed with practical implications for designing and evaluating generalizable RL agents.

Abstract

In recent years, machine learning approaches have made dramatic advances, reaching superhuman performance in Go, Atari, and poker variants. These games, and others before them, have served not only as a testbed but have also helped to push the boundaries of AI research. Continuing this tradition, we examine the tabletop game Dominion and discuss the properties that make it well-suited to serve as a benchmark for the next generation of reinforcement learning (RL) algorithms. We also present the Dominion Online Dataset, a collection of over 2,000,000 games of Dominion played by experienced players on the Dominion Online webserver. Finally, we introduce an RL baseline bot that uses existing techniques to beat common heuristic-based bots, and shows competitive performance against the previously strongest bot, Provincial.
Paper Structure (11 sections, 1 figure, 3 tables)

This paper contains 11 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Dominion
  3. Dominion Online Dataset
  4. Benchmarks
  5. Buy Menus
  6. Provincial
  7. DQN-bot
  8. Conclusion
  9. Appendix
  10. Game Logs
  11. Additional Dominion Agents

Figures (1)

  • Figure 1: (a) Typical game setup on the Dominion Online web server. The basic cards in the blue rectangle are present in every game of Dominion, whereas the 10 kingdom cards in the magenta rectangle are chosen randomly from a set of up to 350 cards. A dominion card has the card cost in the bottom left and the number of copies in the top left. (b) A visualization of the leading strategies Provincial discovered given a random cardset (including cards from later expansions). From left to right, we see 1) the expected win ratio when each of the five strategies plays against each other strategy averaged over 10,000 games, 2) the victory card purchase thresholds, and 3) the buy menus.