marcofavorito / breakout-env

A configurable Breakout environment for reinforcement learning

Breakout-env

A gym like Breakout environment but with more configurable options.

Configurable Options

Option	Description	Type	Range	Default Value
max_step	Max step per episode.	int	0 ~ Inf	10000
lifes	Lifes per episode.	int	0 ~ 9	5
ball_pos	Ball's initial position. [y, x]	[int, int]	-Inf ~ Inf	[100, 40]
ball_speed	Ball's initial velocity. [y, x]	[int, int]	-Inf ~ Inf	[4, 2]
ball_color	Ball's color. (gray scale)	int	0 ~ 255	143
ball_size	Ball's size. [h, w]	[int, int]	1 ~ Inf	[5, 2]
paddle_width	Paddle's width.	int	1 ~ 100	15
paddle_color	Paddle's color. (gray scale)	int	0 ~ 255	143
paddle_speed	Paddle's moving speed.	int	1 ~ Inf	3
bricks_rows	Number of bricks row.	int	0 ~ Inf	6
bricks_color	Row color of bricks.*	list of int	0 ~ 255	[200, 180, 160, 140, 120, 100]
bricks_reward	The reward of bricks.*	list of int	-Inf ~ Inf	[6, 5, 4, 3, 2, 1]

* len(bricks_color) and len(bricks_color) should equal to bricks_rows.

Installation

Install from PyPI:

$ pip install breakout_env

Install from master branch

$ pip install git+https://github.com/SSARCandy/breakout-env.git@master

Example

Interact with environment

from breakout_env import Breakout

# Create Breakout environment with some options.
env = Breakout({
    'lifes': 7,
    'paddle_width': 30,
    'paddle_speed': 5
  })

for ep in range(2):
  obs = env.reset()
  while True:
    # Select random action
    action = random.randint(0, env.actions - 1)
    obs, reward, done, _ = env.step(action)
    print('Episode: {}, Reward: {}, Done: {}'.format(ep, reward, done))
    if done:
      break

Visualize the game

The observation retuned by env is a numpy 2D array, it can be easily visualize using some library like OpenCV or matplotlib.

import cv2
from breakout_env import Breakout

env = Breakout()
env.reset()

while True:
  # Select random action
  action = random.randint(0, env.actions - 1)
  obs, _, _, _ = env.step(action)
  # Show the observation using OpenCV
  cv2.imshow('obs', obs)
  cv2.waitKey(1)
  if done:
    break

Develop Requirement

• python3
• numpy

Reference

• OpenAI Gym
• ALE interface - Breakout
• BreakoutNoFrameskip-v4 video