SECOND-Grasp: Semantic Contact-Guided Dexterous Grasping
Researchers propose SECOND-Grasp (SEmantic CONtact-guided Dexterous Grasping), a unified framework integrating physically stable dexterous grasping with language-guided semantic reasoning. The approach uses vision-language reasoning to generate coarse contact proposals based on object properties, then segments these regions across multiple views for consistency. This enables robotic hands to dynamically adjust grasping strategies for both physical feasibility and semantic understanding, addressing the typical separation of stability and task guidance in robotic manipulation.
Key facts
- SECOND-Grasp stands for SEmantic CONtact-guided Dexterous Grasping
- Framework unifies physically stable grasping and language-guided grasp generation
- Uses vision-language reasoning for contact proposals
- Segments contact regions across multiple views
- Enables dynamic adjustment of grasping strategies
- Addresses separation of stability and semantic guidance
- Published on arXiv with ID 2605.13117
- Announce type is cross
Entities
Institutions
- arXiv