Using PyBoy with Gym

As most people opted to modify the OpenAI Gym that PyBoy used to have, we've decided to remove the Gym Env from the codebase itself and replace it with this example. To use it, copy it into your codebase, and modify it to your needs.

To install the requirements, run: pip install pyboy numpy gymnasium.

Thanks to Nicole Faye for providing the basis for this sample code. If you have questions or improvements to the code, come and tell us on Discord.

# Adopted from https://github.com/NicoleFaye/PyBoy/blob/rl-test/PokemonPinballEnv.py
import gymnasium as gym
from gymnasium import spaces
import numpy as np
from pyboy import PyBoy

actions = ['','a', 'b', 'left', 'right', 'up', 'down', 'start', 'select']

matrix_shape = (16, 20)
game_area_observation_space = spaces.Box(low=0, high=255, shape=matrix_shape, dtype=np.uint8)

class GenericPyBoyEnv(gym.Env):

    def __init__(self, pyboy, debug=False):
        super().__init__()
        self.pyboy = pyboy
        self._fitness=0
        self._previous_fitness=0
        self.debug = debug

        if not self.debug:
            self.pyboy.set_emulation_speed(0)

        self.action_space = spaces.Discrete(len(actions))
        self.observation_space = game_area_observation_space

        self.pyboy.game_wrapper.start_game()

    def step(self, action):
        assert self.action_space.contains(action), "%r (%s) invalid" % (action, type(action))

        # Move the agent
        if action == 0:
            pass
        else:
            self.pyboy.button(actions[action])

        # Consider disabling renderer when not needed to improve speed:
        # self.pyboy.tick(1, False)
        self.pyboy.tick(1)

        done = self.pyboy.game_wrapper.game_over

        self._calculate_fitness()
        reward=self._fitness-self._previous_fitness

        observation=self.pyboy.game_area()
        info = {}
        truncated = False

        return observation, reward, done, truncated, info

    def _calculate_fitness(self):
        self._previous_fitness=self._fitness

        # NOTE: Only some game wrappers will provide a score
        # If not, you'll have to investigate how to score the game yourself
        self._fitness=self.pyboy.game_wrapper.score

    def reset(self, **kwargs):
        self.pyboy.game_wrapper.reset_game()
        self._fitness=0
        self._previous_fitness=0

        observation=self.pyboy.game_area()
        info = {}
        return observation, info

    def render(self, mode='human'):
        pass

    def close(self):
        self.pyboy.stop()