Snake for Gym

Snake environment for OpenAI Gym, written in Python 3.

Dependencies

Gym
Pygame

Run it like a Gym environment:

from snake.env import SnakeEnv

env = SnakeEnv()

for episode in range(10):
    observation = env.reset()
    while True:
        env.render()
        action = env.action_space.sample()
        observation, reward, done, info = env.step(action)
        if done:
            break

Play with it yourself!

env = SnakeEnv()
env.interactive()

SnakeEnv

Initialization parameters

state_size: the game is composed of nxn blocks, and size specifies n.
snake_size: size of the snake
reward_eat: reward for eating a dot/fruit.
reward_win: reward for winning (max score).
reward_lose: reward for losing.
reward_time: time discount reward
fps: frames per second (for rendering).
unit: specifies how big each block is (for rendering).

Default: state_size=20, snake_size=3, reward_eat=1, reward_win=10, reward_lose=-1, reward_time=-0.01, fps=10, unit=10

By default the snake starts at the midpoint and goes upwards.

Actions

0: Do nothing

1: Up

2: Down

3: Left

4: Right

Observation

Returns an nxn numpy array.

Empty blocks are represented by 0's
Snake body is represented by 1's
The reward is represented by 2.

Info

env.step also returns a dictionary named 'info'.

If the game has ended, (i.e. done == True), then you can inspect info

if done: print(info['cause of death'])

To examine the cause of death. It takes on one of two string values: 'out of bounds' or 'self collision'.

Alternatives

PyGame Learning Environment also has a snake environment with other games as well.

YuriyGuts's snake for AI has examples and GPU support.