AIBullisharXiv – CS AI · 7h ago7/10
🧠
LUCID: Learning Embodiment-Agnostic Intent Models from Unstructured Human Videos for Scalable Dexterous Robot Skill Acquisition
LUCID is a machine learning framework that learns robot manipulation skills from unstructured internet videos and human demonstrations, then transfers this knowledge to different robot embodiments through a shared intent model. The approach eliminates the need for expensive, embodiment-specific robot training data and demonstrates zero-shot transfer capabilities across multiple real-world tasks.