snake-gym-grid

snake-gym

This package is my own implementation on Snake game with Gym integration.

Usage

import gym
import snake_gym_grid

env = gym.make("snake-gym-10x20-normal-v0")
env.reset()

done = False # whether the snake dies

"""
action ranges from 0 to 3.
0 - LEFT
1 - RIGHT
2 - UP
3 - DOWN
"""
action = ...

observation = env.reset()

while not done:
    env.render()
    observation, reward, done, info = env.step(action)
env.close()

Output at each time step

• observation: Pixel image of the game
• reward: 1.0 when a food is eaten, -1.0 if snake dies, 0.0 otherwise
• done: whether the snake dies
• truncate: whether the game is running for more than 100000 time steps
• info: an empty dictionary for now

Env registration

• snake-gym-grid-10x20-normal-v0: Snake game with 10 x 20 board and 500 x 900 observation
• snake-gym-grid-10x20-tiny-v0: Snake game with 10 x 20 board and 80 x 80 observation
• snake-gym-grid-10x20-1d-v0: Snake game with 10 x 20 board and summarized observation. Observation in this case is vector [dangerRight, dangerLeft, dangerBelow, dangerAbove, foodBelow, foodAbove, foodRight, foodLeft].

Note:

• danger<Direction> = neighbor block of snake head in the <Direction> is occupied by a snake body or is out of game board.

• food<Direction> = is the food in the <Direction> of the snake head?

• For any other environment, observation is the raw image from the game view. For tiny, the size is 80 x 80 and for normal, it's 500 x 900.

Customize environment

Instead of using registered environment, one can utilize the built-in SnakeGymGrid class.

# this import is kinda ugly but I have no choice but to follow the gym standard
from snake_gym_grid.envs.snake_gym_grid import SnakeGymGrid

env = SnakeGymGrid(width=..., height=..., n_rows=..., n_cols=...)
...