Garment & Deformable-Object Manipulation Data
Why is garment-manipulation data hard and valuable?
Deformable objects like cloth have near-infinite configurations, self-occlude and deform unpredictably when grasped, making garment manipulation one of the hardest open problems in robotics. High-quality demonstrations from skilled tailors are scarce and valuable - and apply to a large apparel, laundry and soft-goods market.
What nxted captures
Recorded from skilled tailors and textile workers, often with a second reference angle to handle occlusion.
- ›Aligning, folding and pressing fabric
- ›Hand and machine stitching, pattern handling
- ›Bimanual coordination across sub-steps
- ›Hand pose and action segmentation with success labels
- ›Diversity of fabrics, garments and conditions
Why deformable-object data is scarce
Rigid-object methods do not transfer to cloth, and simulation struggles with fabric dynamics, so real human demonstrations are the practical path. Tailors bring decades of dexterity that gig crowds cannot match.
How nxted delivers it
Robotics-ready in LeRobot, RLDS and HDF5 with metadata, a dataset card and a Data Trust Pack. Start with a Physical AI Test Kit.
FAQ
Why is cloth hard for robots?
Fabric has near-infinite configurations, self-occludes and deforms as you grasp it, so methods built for rigid objects do not transfer. Real human demonstrations are the most practical training signal.
Who provides garment demonstrations?
Skilled tailors and textile workers, consented and fairly paid, recorded egocentrically (often multi-view) with hand-pose and action labels.
What formats?
LeRobot, RLDS and HDF5 with metadata, dataset card and QA report.