Lifelong Learning of Factored Policies via Policy Gradients