Solving robust MDPs as a sequence of static RL problems