Learning Dexterous Grasping from Sparse Taxonomy Guidance

Researchers developed GRIT, a two-stage AI framework that learns dexterous robotic grasping from sparse taxonomy guidance, achieving 87.9% success rate. The system first predicts grasp specifications from scene context, then generates finger motions while preserving intended grasp structure, improving generalization to novel objects.

• →GRIT framework combines taxonomy-based grasp prediction with continuous finger motion generation for dexterous manipulation.
• →The system achieved 87.9% overall success rate and demonstrated improved generalization to novel objects compared to baselines.
• →Certain grasp taxonomies prove more effective for specific object geometries, enabling better task performance.
• →Real-world experiments show the system allows controllable grasp strategy adjustment through high-level taxonomy selection.
• →The approach addresses the impracticality of specifying dense pose targets while maintaining user controllability over pure reinforcement learning.