Reproduce “MetaFold: Language-Guided Multi-Category Garment Folding Framework via Trajectory Generation and Foundation Model” 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
Haonan Chen, Junxiao Li, Ruihai Wu, Yiwei Liu, Yiwen Hou, Zhixuan Xu, Jingxiang Guo, Chongkai Gao and 4 more
MetaFold provides a framework for language-guided multi-category garment folding in robotics by disentangling high-level trajectory generation from low-level action prediction. The system achieved up to 94% success rates in real-world experiments and outperformed existing baselines in handling diverse garments and language-conditioned tasks.
Key insights
- Disentangling high-level task planning (trajectory generation) from low-level action prediction significantly enhances robustness and generalization in robotic manipulation of deformable objects.
- Utilizing complete point cloud trajectories for planning, rather than sparse keypoint representations, provides a more comprehensive spatial and temporal understanding crucial for precise garment deformation.
- Integrating a large language model (LLaMA) for semantic feature extraction enables effective interpretation of diverse user instructions and open-vocabulary generalization for complex folding tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.