Abstract: Cloth is an important material to model and simulate correctly, both in computer graphics and other industrial applications. The commonly used models for cloth in computer graphics typically approximate the cloth as an elastic sheet with linear isotropic behavior inspired by the construction of woven fabrics. However, they do a poor job of predicting the behavior of knits, which are driven by the complex interactions of yarn loops pulled through each other. This thesis presents a yarn-based model for cloth where yarns in the fabric are explicitly modeled as inextensible but flexible spline curves. Yarn dynamics are dictated by both energy terms and hard constraints, while friction interactions, a critical component of correct yarn behavior, are approximated using a velocity filter that penalizes locally non-rigid motion. Qualitative comparison of the model to observed deformations of hand-knitted samples in the laboratory showed that the model predicts key mechanical properties of different knits. Since this model is slower than sheet-based approaches, further work looked at accelerating the model through both localized rigidification and adaptive contact linearization. In localized rigidification, regions of the cloth behaving almost rigidly are simulated using a cheaper model which avoids many of the expensive force computations. For adaptive contact linearization, yarn contacts are grouped into contact sets, with the associated contact force computed exactly at one timestep, and then approximated on subsequent steps via linearization in a rotated reference frame for nearby geometric configurations. Finally, additional work looked at the problem of creating initial yarn geometry for subsequent simulation as a yarn-based model.

SIGGRAPH 2010 Fast Forward Video:

Abstract: Yarn-based cloth simulation can improve visual quality but at high computational costs due to the reliance on numerous persistent yarn-yarn contacts to generate material behavior. Finding so many contacts in densely interlinked geometry is a pathological case for traditional collision detection, and the sheer number of contact interactions makes contact processing the simulation bottleneck. In this paper, we propose a method for approximating penalty-based contact forces in yarn-yarn collisions by computing the exact contact response at one time step, then using a rotated linear force model to approximate forces in nearby deformed configurations. Because contacts internal to the cloth exhibit good temporal coherence, sufficient accuracy can be obtained with infrequent updates to the approximation, which are done adaptively in space and time. Furthermore, by tracking contact models we reduce the time to detect new contacts. The end result is a 7- to 9-fold speedup in contact processing and a 4- to 5-fold overall speedup, enabling simulation of character-scale garments.

Other Media:

• Supplemental Video: [Youtube] [MOV, 217MB]
• CAF Papers Video Trailer: [Youtube] [MOV, 133MB]
• Hi res images:
  Sweater [2900x2900, 760KB]
  Afghan [3700x2900, 1.5MB]
  Sack [2500x2500, 1.1MB] (model updates highlighted in red)

Abstract: Knitted fabric is widely used in clothing because of its unique and stretchy behavior, which is fundamentally different from the behavior of woven cloth. The properties of knits come from the nonlinear, three-dimensional kinematics of long, inter-looping yarns, and despite significant advances in cloth animation we still do not know how to simulate knitted fabric faithfully. Existing cloth simulators mainly adopt elastic-sheet mechanical models inspired by woven materials, focusing less on the model itself than on important simulation challenges such as efficiency, stability, and robustness. We define a new computational model for knits in terms of the motion of yarns, rather than the motion of a sheet. Each yarn is modeled as an inextensible, yet otherwise flexible, B-spline tube. To simulate complex knitted garments, we propose an implicit-explicit integrator, with yarn inextensibility constraints imposed using efficient projections. Friction among yarns is approximated using rigid-body velocity filters, and key yarn-yarn interactions are mediated by stiff penalty forces. Our results show that this simple model predicts the key mechanical properties of different knits, as demonstrated by qualitative comparisons to observed deformations of actual samples in the laboratory, and that the simulator can scale up to substantial animations with complex dynamic motion.

Other Media

• Supplemental Video: [Youtube] [MOV, 120MB]
• Cover Image from Conference Proceedings [2346x2580, 650KB]
• Computer Animation Festival movie [MOV, 10MB]

Coming soon!