Forgot to mention that it is learning from raw pixels!! No other info is passed to the agent, except the screen frames :)