Inductive Policy Selection for First-Order MDPs