VLP: Vision-Language Preference Learning for Embodied Manipulation