About #

RootAI is a non-neural network artificial intelligence ( AI ) system that is capable of playing the boardgame “Root.” It is the final project of my bachelor’s degree. We experimentd by creating various different agents, then running multiple battles between them. One best AI agent for each faction were found. They were able to play as good as an experienced human* (according to Reddit and our own experience), while using reasonable* thinking time.

RootAI is set in the scenario of a 2 faction game — Marquise de Cat vs Eyrie Dynasties.

• Download & Play(?) / Run RootAI on GitHub
• See RootAI’s source code on GitHub

Running the game #

Prerequisite #

• knowledge of Root boardgame’s rules (official Law of Root pdf)

Download #

• download RootAI.zip from GitHub releases
• extract the zip file

Run #

• double click RootAI.exe

Config #

config file is in /config/config.yml, change to match your needs.

• Eyrie is set to be the MCTS AI agent
• Marquise is the player
• Eyrie starts first

Controls #

• Press UP/DOWN — change selected action
• Press R — random action
• Press RETURN/SPACE — execute selected action
• Hold F — continuously run player
• Hold A — continuously run agent
• Press N in game-end state — new game
• Press Q in game-end state — quit game
• Press O — print_game_state
• Press C — new_game_from_current_game_state

Development #

• Graphics Engine - Pygame
• Programming Language - Python

Breakdown of the program #

RootMinimal #

RootMinimal is the game environment with the following features:

• Simulate the current state of the game
• Generate all possible actions (“legal actions”) for the current player at the current state of the game
• Change state according to the selected action Certain mechanics that are not stated in the rulebook are intepreted according to the Root digital game.

RootTrainer #

RootTrainer is a framework allows humans and AI agents to interact with RootMinimal’s environment.

The AI agents are implemented using these methods:

• Random decision
• Monte Carlo Tree Search (MCTS) algorithm with multiple variants
  • One Depth MCTS
  • General MCTS with various settings

Agent: Random decision #

It just picks a random action.

Agent: Monte Carlo Tree Search (MCTS) #

The MCTS algorithm maximize the probability of winning by simulating rollouts. Basically, it branch out to each action, simulate randomly (multiple times) until a set condition, and propagates the “goodness” of the game state that it reached back to the root node.

Then it picks the best action by looking at the results.

Read more at Wikipedia: Monte Carlo tree search.

Gallery #