ZOZO's Contact Solver (for physics-based simulations)
6 months ago
- #physics-simulation
- #GPU-computing
- #contact-solver
- A contact solver for physics-based simulations involving shells, solids, and rods, developed by ZOZO.
- Features include robust penetration-free contact resolutions, scalability up to 150M contacts, and GPU efficiency in single precision.
- The solver ensures cloth inextensibility beyond strict bounds (e.g., 1%) and is driven by the Finite Element Method for physical accuracy.
- Includes GitHub Actions for stress testing, Docker for out-of-the-box functionality, and JupyterLab for immediate example execution.
- Comprehensive Python API documentation and cloud-ready deployment on major platforms like AWS, Google Compute Engine, and vast.ai.
- Recent updates include codebase refactoring, simulation state save/load functionality, and addition of multiple examples.
- Published in ACM Transactions on Graphics (TOG) Vol.43, No.6, with supplementary materials and presentation videos available.
- Designed for cost-effective cloud deployment, leveraging NVIDIA L4 GPUs for scalable performance.
- GitHub Actions ensure reliability by running tests 10 times with jittered object positions to simulate varied conditions.
- Acknowledgements to ZOZO, Inc. for supporting the project's release and development.