Physically based cloth simulation is widely used to generate realistic animations for films, games, and virtual environments. However, achieving artist-directed cloth motion often requires costly trial-and-error adjustment of simulation parameters, making the process increasingly impractical for high-resolution simulations. This thesis presents a coarse-to-fine framework for artist-directed cloth animation that combines differentiable inverse simulation with Progressive Dynamics. Artistic intent is formulated as a differentiable objective over the simulated cloth trajectory, enabling user-defined goals, such as target poses or motions, to be optimized directly through gradient-based methods.

To improve computational efficiency, inverse optimization is performed over a hierarchy of cloth discretizations. Optimization begins on a coarse representation of the cloth, where large-scale motion can be recovered efficiently, before the optimized state and control parameters are progressively transferred to finer resolutions to reconstruct high-frequency geometric detail. By integrating Progressive Dynamics with differentiable inverse simulation, the proposed framework reduces the computational cost of inverse optimization while preserving physically plausible cloth behavior. Experimental results demonstrate improved optimization efficiency compared to direct high-resolution inverse simulation, providing a more practical and intuitive workflow for artist-directed cloth animation.

44 pages

Thesis Committee:
Minchen Li (Chair)
James McCann

Jignesh Patel, Interim Head, Computer Science Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu