In this followup work, we want to learn the subgoal structure as representations of the policy sketches language on increasingly more difficult and diverse goal descriptions. The goal is to show that we can reuse previously acquired knowledge to changes in dynamic of the domain.