How to Peel with a Knife: Aligning Fine-Grained Manipulation with Human Preference

Researchers developed a two-stage learning framework enabling robots to perform complex manipulation tasks like food peeling with over 90% success rates. The system combines force-aware imitation learning with human preference-based refinement, achieving strong generalization across different produce types using only 50-200 training examples.

• →New AI framework enables robots to master complex manipulation tasks like peeling fruits and vegetables with 90%+ success rates.
• →Two-stage approach combines initial imitation learning with human preference-based policy refinement for subjective task quality.
• →System requires only 50-200 training trajectories and shows strong zero-shot generalization to unseen objects.
• →Performance improved by up to 40% through preference-based finetuning using learned reward models.
• →Framework addresses contact-rich tasks with implicit success criteria that are difficult to quantify objectively.