Abstract
Techniques are directed to a self-supervised neural network simulator configured to generate 3D hair animations for interactive platforms such as virtual reality and video games. By replacing computationally expensive physics solvers with explicit particle chain representations and a discrete volumetric mass-spring topology in 3D Euclidean space, the framework maps skeletal motion, character body shape, and rest-state hairstyle geometry into per-strand hair deformations. The system architecture integrates a Latent Groom Transformer to resolve global strand interactions via cross-attention mechanisms, an Identity Encoder to capture body proportions, a recurrent Gated Recurrent Unit Motion Encoder to continuously track momentum history such as drag and follow-through, and a Modulated Neural Simulator that fuses dynamic movement with static character parameters. Crucially, because static hairstyle and body shape features are encoded strictly once per character, per-frame runtime execution is isolated to lightweight updates, achieving real-time animation speeds while preserving macroscopic hair volume and natural physical dynamics.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Park, Hyo Joon; Zhou, Yuxiao; Bertiche Argila, Hugo; and Chai, Menglei, "Neural Hair Simulator", Technical Disclosure Commons, (November 09, 2026)
https://www.tdcommons.org/dpubs_series/12837