RGBench Cloth Sim-to-Real (v1) contains nine carefully captured garments, each manipulated with three bimanual actions (fling, fold, grasp). The dataset provides real-world ground truth point clouds for evaluating the sim-to-real gap of cloth simulators. It was released by author hwk0809 as part of the AAAI 2026 paper Real Garment Benchmark (RGBench).
Use Cases
- Evaluate cloth simulator performance based on real-world ground truth point clouds.
- Benchmark sim-to-real transfer for robotic bimanual manipulation tasks based on the described fling, fold, and grasp actions.
- Train or validate machine learning models for cloth state estimation based on the provided point cloud data.
Strengths
- Includes nine distinct garments, each captured with three specific manipulation actions.
- Provides real-world ground truth point clouds, a key feature for sim-to-real evaluation.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- hwk0809
- Collection Method
- Carefully captured garments with real-world ground truth point clouds.
- Freshness
- Last updated 2026-05-16 09:30:21; freshness should be verified.