pongram.py

import numpy as np
import gym

# Pong env
env = gym.make('Pong-ram-v0')
# Reset env
env.reset()
# Initialize previous obs
prev_obs = np.zeros((128,))
# Render and test
while True:
    # Render
    env.render()
    # Sample an action from action space [0-5]
    action = env.action_space.sample()
    # Take an action
    observation, reward, done, info = env.step(action)
    # Difference between previous obs and current obs
    move = prev_obs - observation
    # Current obs will become the previous obs
    prev_obs = observation
    # Break once done
    if done:
        break