Gymboy 🤖
Gameboy (Color) Environments in Gymnasium

Gymboy supports a range of different RL environments from the Game Boy Color using the Gymnasium API.

Implemented Environments 🌍

Environment Name	Python Source
Kirby-Dream-Land-1-minimal-image-v1	Click
Kirby-Dream-Land-1-full-image-v1	Click
Pokemon-Blue-minimal-image-v1	Click
Pokemon-Blue-full-image-v1	Click
Pokemon-Gold-minimal-image-v1	Click
Pokemon-Gold-full-image-v1	Click
Pokemon-Red-minimal-image-v1	Click
Pokemon-Red-full-image-v1	Click
Pokemon-Silver-minimal-image-v1	Click
Pokemon-Silver-full-image-v1	Click
Pokemon-Yellow-minimal-image-v1	Click
Pokemon-Yellow-full-image-v1	Click
Super-Mario-Land-1-minimal-image-v1	Click
Super-Mario-Land-1-full-image-v1	Click
Tetris-minimal-image-v1	Click
Tetris-full-image-v1	Click

Installation ⚙️

Install the package via pip:

pip install gymboy

⚠️ Important: Gymboy requires specific ROM files to function properly. Make sure you have the necessary ROMs available before using any of the environments.

Usage 🚀

Here's a quick example of how to use a gymboy environment:

import numpy as np

import gymboy

# Create the environment
env = gymboy.make(
    env_id="Pokemon-Blue-full-image-v1",
    rom_path="./resources/roms/pokemon/gen_1/pokemon_blue.gb",
    init_state_path="./resources/states/pokemon/gen_1/pokemon_blue_after_intro.state",
)
num_steps = 1000

# Reset the environment
observation, info = env.reset()
for i in range(num_steps):
    # Sample a random action
    action = env.action_space.sample()

    # Perform the action
    observation, reward, terminated, truncated, info = env.step(action)
    done = np.logical_or(terminated, truncated)

    if done:
        # Case: Environment has terminated
        break

# Close the environment
env.close()

You can also create multiple instances of the environment running in parallel:

import numpy as np

import gymboy

# Create the environments
envs = gymboy.make_vec(
    env_id="Pokemon-Blue-full-image-v1",
    num_envs=2,
    rom_path="./resources/roms/pokemon/gen_1/pokemon_blue.gb",
    init_state_path="./resources/states/pokemon/gen_1/pokemon_blue_after_intro.state",
)
num_steps = 1000

# Reset the environments
observations, infos = envs.reset()
for i in range(num_steps):
    # Sample random actions
    actions = envs.action_space.sample()

    # Perform the actions
    observations, rewards, terminated, truncated, infos = envs.step(actions)
    dones = np.logical_or(terminated, truncated)
    # No need to check for dones — environments auto-reset internally

# Close the environments
envs.close()

Development 🔧

Contributions are welcome!

Please fork the repository and submit a pull request.

Make sure to follow the coding standards and write tests for any new features or bug fixes.