World-action models (WAM) predict future states to guide robot actions, enabling learning from both action-free video an…

Robot learning increasingly depends on broad and diverse demonstrations, yet collecting robot data remains expensive and…