Amazons

The Game of the Amazons is a two-player abstract strategy game invented in 1988 by Walter Zamkauskas of Argentina. The game is played by moving pieces and blocking the opponents from squares, and the last player able to move is the winner.

Roadmap

• Implement the Amazons game in C++ and export to Python using pybind11.
• Creating a custom Pettingzoo environment.
• Design a simple GUI.
• Train AI agents to play Amazons. (This may not be achievable.)

amazons_env

amazons_env.raw_env inherits from pettingzoo.AECEnv. As action masking is not implemented, it includes a random_action method to generate a random (valid) action. Note that this method relies on brute force and can be computationally expensive.

import amazons_env
env = amazons_env.env()
env.reset()

for agent in env.agent_iter():
    observation, reward, termination, truncation, info = env.last()

    if termination or truncation:
        action = None
    else:
        action = env.unwrapped.random_action()

    env.step(action)

env.close()

• agent can take on values of either "black_0" or "white_0".
• observation is structured as {'observation': np.ndarray}. The array is Boolean and has a shape of [N, N, 3], encoding the player's stones, the opponent's stones, and the arrows.
• action is expected to be a numpy array with a shape of [6,], representing $(x_0, y_0, x_1, y_1, x_2, y_2)$. Each coordinate should fall within the range [0, N).

For the moment, amazons_env.raw_env has the following limitations:

• Lack of support for any render_mode.
• Absence of an action-masking mechanism.

Requirements

• CMake

Reference

• https://en.wikipedia.org/wiki/Game_of_the_Amazons

• https://ieeexplore.ieee.org/document/9534274