Cloth Real Robot Rollouts: Synchronized Point Clouds for Folding Tasks

Real-world point cloud recordings of a folding and unfolding cloth, used to evaluate state-estimation models from UniClothDiff. Each rollout contains two synchronized point cloud streams derived from an Intel RealSense depth camera and a monocular RGB image. The dataset was authored by Cloth-splatters and last updated on Hugging Face in May 2026.

Use Cases

• Benchmark state-estimation models based on the synchronized point cloud streams.
• Train monocular depth estimation models based on the RGB-to-depth data pairs.
• Evaluate robotic manipulation policies for cloth folding and unfolding tasks.
• Develop sim-to-real transfer methods for deformable object handling.

Strengths

• Contains two synchronized data streams (camera depth and monocular estimation) for direct comparison.
• Positions are stored with a defined scale (int16 in units of 0.1 mm) for precise measurement.
• Designed for a specific, concrete evaluation task (assessing UniClothDiff models).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description references a full page for details like quantisation error, requiring external navigation.

Provenance

Source

Cloth-splatters

Collection Method

Recordings from a real robot setup using an Intel RealSense depth camera.

Freshness

Last updated 2026-05-11 12:14:06.