Abstract: Learning control policies from high-dimensional robot demonstrations remains challenging due to the need for large datasets, limited generalization, and high model complexity. These ...