Learning Robust Helpful Behaviors in Two-Player Cooperative Atari Environments