To run snake game without gym environment, you can look into folder multiagent-cooperative env. Open jupyter notebook from same and run the sample code.

• Install OpenAI Gym and Baselines
• Install IPython and numpy via pip or conda.

To add the environment, do

cd gym-np-snake
pip install -e .

The gym environment is only supporting single agent setup. However the Snake Trainer and SnakeEnv as listed in parent directory is designed to work for MultiAgent Cooperative setup.

To use the snakegame as gym-environment, use the bellow code sample :

import gym;
import gym_np_snake;
env = gym.make('SnakeNp-v0')
from time import sleep
reward=0;

while reward>-0.5:    
    ob,reward,dones,info = env.step(0)
    env.render(mode='human')
    sleep(0.1)

Reward is between -1 and 1. -1 for collision and 1 for eating the food.
Observation is 10x10x1 image output of encoded display. Mapping is as follows

Info contains the state information such 4 element tuple with following values (direction/4, normalized x distance, normalized y distance, distance of head from food divided by length+width) Interpretation of direction for the game is as follow :

The actions are Discrete(3) , which means, either a snake can go left, go straight or go right. This is as follow :

To have an overview of approach. The details are going to be posted on youtube. Bellow is the link of introductory video for the same.