Reproduce “Cross-Category Functional Grasp Transfer” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Rina Wu, Tianqiang Zhu, Xiangbo Lin, Yi Sun

Researchers from Dalian University of Technology developed a method for cross-category functional grasp transfer in multi-fingered robots by leveraging object similarity relationships and a knowledge graph. This approach achieved competitive grasp success rates for unseen objects and significantly faster inference times compared to prior within-category methods.

Key insights

  • Modeling human-like similarity relationships (shape, part layout, and functional grasp type) between objects is effective for bridging differences and inferring functional grasps for novel categories.
  • A Graph Attention Network (GAT) trained on these similarity relationships within a knowledge graph can successfully predict hand-agnostic 'touch codes' for objects in unseen categories.
  • Combining semantic touch code transfer with an optimization-based synthesis module enables generalizable and efficient functional grasp generation for complex dexterous hands.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.