There are many goals, how do the agent select the next goal to complete? It is not clear in the paper.