Self-supervised Learning of Latent Space Dynamics

Authors:
  • YLYue Li
    Yue Li
  • GLGene Wei-Chin Lin
    Gene Wei-Chin Lin
  • ELEgor Larionov
    Egor Larionov
  • ABAljaz Bozic
    Aljaz Bozic
  • DRDoug Roble
    Doug Roble
  • LKLadislav Kavan
    Ladislav Kavan
  • SCStelian Coros
    Stelian Coros
  • BTBernhard Thomaszewski
    Bernhard Thomaszewski
  • TSTuur Stuyck
    Tuur Stuyck
  • HCHsiao-Yu Chen
    Hsiao-Yu Chen

Abstract
Modeling the dynamic behavior of deformable objects is crucial for creating realistic digital worlds. While conventional simulations produce high-quality motions, their computational costs are often prohibitive. Subspace simulation techniques address this challenge by restricting deformations to a lower-dimensional space, improving performance while maintaining visually compelling results. However, even subspace methods struggle to meet the stringent performance demands of portable devices such as virtual reality headsets and mobile platforms. To overcome this limitation, we introduce a novel subspace simulation framework powered by a neural latent-space integrator. Our approach leverages self-supervised learning to enhance inference stability and generalization. By operating entirely within latent space, our method eliminates the need for full-space computations, resulting in a highly efficient method well-suited for deployment on portable devices. We demonstrate the effectiveness of our approach on challenging examples involving rods, shells, and solids, showcasing its versatility and potential for widespread adoption.

This publication has 23 references indexed in Scilit: