Snake game Custom environment in Python using OpenAI Gym

By Rhushabh Gedam

Snake-gym is a custom environment of the classic snake game using OpenAI Gym. It is meant to test different Reinforcement Learning Agents. It has a "human" mode to serve as a Single Player game.

Installation:

$> pip install -r requirements.txt

How to run:

$> python3 main.py [--mode ] [--episodes ]

where mode = "bot" or "human" [default = "bot"],
num_episodes = number of games to simulate [default = 30]

Controller Functions:

• reset(): reset env to start state
• step(action): return next-state and reward based on action
• render(): render env on screen
• play(max_moves): simulate a game with maximum moves = max_moves
• close(): quit the env

Rewards and Penalties:

-1000 if snake dies
+10 if snake ate food
-1 if snake moves 1 block

Game Details:

A sample Agent class is defined to demonstrate the Agent-Environment interaction. Average Scores of all the games played is displayed at the end. The grid dimensions are 20x20.

Snake is allowed to move in 3 directions w.r.t. its body orientation:
LEFT (left arrow key), STRAIGHT (up arrow key), RIGHT (right arrow key)

The state of the environment at any instant is modelled as a complete 20x20 grid with the following notation:
0 -> empty cell
1 -> snake head + body
2 -> food

Game terminates when the snake hits his head on the wall or tries to eat itself.