Abstract
In robotic cloth manipulation, the lack of standardized object descriptions hinders reproducibility and performance analysis. This paper introduces a non-destructive, low-cost framework to quantify essential physical and mechanical properties—such as stiffness, elasticity, and friction—specifically tailored for robotics research. By applying this framework to existing cloth datasets, the authors demonstrate that these underlying material properties significantly impact the success of common manipulation tasks like folding, dragging, and lifting.
Introduction
Unlike rigid body robotics, where mass, dimensions, and friction coefficients are typically well-defined, cloth manipulation research often relies on vague descriptors like "a T-shirt" or "a towel." This lack of standardization is a significant barrier to reproducibility; two garments of the same category can exhibit vastly different behaviors due to variations in material composition and construction. While the textile industry utilizes standardized tests like KES or FAST, these are often destructive or require specialized, expensive equipment. This work addresses the need for a non-destructive characterization framework that allows roboticists to precisely label cloth "actors" and understand how material properties influence manipulation outcomes.
Characterization Framework
The proposed framework adapts textile engineering principles for a standard robotics laboratory environment, focusing on non-destructive methods.
Physical Properties
The framework mandates standardized reporting of size (using specific reference lines), weight, shape category, color, material, and construction technique (e.g., woven vs. knitted).
Mechanical Properties
The "Big Three" mechanical properties are measured using simple tools:
- Stiffness: Inspired by the Cusick drape test, the cloth is placed on a circular plate. A zenithal (top-down) RGB camera captures the draped area, and a stiffness formula is calculated based on the ratio of the projected area to the initial area.
- Elasticity: Measured by recording the elongation ($l_f - l_i / l_i$) of the cloth when pulled with a standard 0.5kg luggage scale.
- Friction: Determined via an inclined plane test. The cloth is placed on a surface, and the angle ($\theta$) is increased until sliding occurs, allowing the derivation of the friction coefficient ($\mu = \tan(\theta)$).
Dataset Analysis
The authors evaluated three prominent cloth datasets: the Elastic Object Set (EOS), the Household Cloth Object Set (HCOS), and the Dressing Object Set (DOS). By visualizing these datasets on a radar chart, the researchers highlighted the unique characteristics and "diversity" of each set. For instance, HCOS is characterized by larger, heavier items, while EOS provides the greatest variation in mechanical properties.
Key Insights and Results
Experimental testing of five primitives—Lift, Drag, Fold, Pull, and Push—revealed several critical insights:
- Stiffness Dominance: Stiffness was found to be the primary property influencing shape retention across most tasks. For example, Sample A (highest stiffness) achieved the best results in Lift and Fold tasks.
- Property Interdependence: Mechanical properties are often linked to physical ones; in the HCOS set, high stiffness was frequently a byproduct of larger size and weight rather than weave density alone.
- Contextual Importance: While stiffness dictates shape retention, elasticity becomes critical under stress (pulling), and friction is the deciding factor during surface interactions like pushing or dragging.
- Success Metrics: Using a "Final Ratio" (FR) metric for shape retention, the study showed that "floppy" cloths (low stiffness) are objectively harder to manipulate predictably, resulting in lower FR scores.
Conclusion and Future Work
This framework provides a vital step toward standardizing cloth manipulation research. However, it currently treats cloth as a homogeneous material, ignoring local inhomogeneities like buttons, zippers, or hems. Future research will focus on incorporating local measurements and testing more complex, non-quasi-static manipulations. The ultimate goal is to establish these "cloth descriptors" as standard reporting requirements in the robotics community.
Summary updated