Applying Policy Gradient Methods To Open-Ended Domains