Factored Policy Gradients: Leveraging Structure for Efficient Learning in MOMDPs

Policy gradient methods can solve complex tasks but often fail when the dimensionality of the action-space or objective multiplicity grow very large.