Tag
This paper analyzes the capability separation between world-model policy learning and imitated world-action models, demonstrating that world-model learning differs in its decision rule and information requirements under certain conditions.