DiffCloth: Differentiable Cloth Simulation with Dry Frictional Contact

TL;DR

This paper introduces DiffCloth, a differentiable cloth simulation framework based on Projective Dynamics with dry frictional contact, enabling faster and more accurate cloth-related applications.

Contribution

It presents a novel, fast method for deriving gradients in PD-based cloth simulation with dry frictional contact, enhancing various downstream applications.

Findings

Significant speedup in applications using gradient information

Effective in system identification and trajectory optimization

Improves real-to-sim transfer accuracy

Abstract

Cloth simulation has wide applications in computer animation, garment design, and robot-assisted dressing. This work presents a differentiable cloth simulator whose additional gradient information facilitates cloth-related applications. Our differentiable simulator extends a state-of-the-art cloth simulator based on Projective Dynamics (PD) and with dry frictional contact. We draw inspiration from previous work to propose a fast and novel method for deriving gradients in PD-based cloth simulation with dry frictional contact. Furthermore, we conduct a comprehensive analysis and evaluation of the usefulness of gradients in contact-rich cloth simulation. Finally, we demonstrate the efficacy of our simulator in a number of downstream applications, including system identification, trajectory optimization for assisted dressing, closed-loop control, inverse design, and real-to-sim transfer. We observe a substantial speedup obtained from using our gradient information in solving most of these applications.