Reproduce “TRTM: Template-based Reconstruction and Target-oriented Manipulation of Crumpled Cloths” 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
Wenbo Wang, Gen Li, Miguel Zamora, Stelian Coros
This work introduces TRTM, a system that reconstructs the full 3D mesh of crumpled cloths from a single top-view depth image using a Graph Neural Network, enabling precise, target-oriented robotic manipulation. The system achieved 97.6% coverage for dual-arm flattening, outperforming a previous agent, and demonstrated high silhouette similarity for targeted folding tasks.
Key insights
- Explicit 3D mesh reconstruction provides a richer geometric understanding of a cloth's state, facilitating more precise and interpretable robotic manipulation actions.
- A GNN architecture, inspired by human body reconstruction, can effectively model the complex deformations and self-occlusions of crumpled cloths.
- Robust sim-real registration and synthetic data generation are effective for training advanced deformable object reconstruction models, reducing the need for costly real-world 3D labeling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.