@pieper presented an early integration of NVIDIA Warp into 3D Slicer. Warp is a Python/CUDA-based simulation framework that supports CPU and GPU execution and integrates efficiently with PyTorch. Notable features include:
All simulations are differentiable by default.
Integration with Slicer is performed using NumPy arrays.
Warp is being refactored into a standalone library called Newton Physics, which will continue using Warp under the hood.
Licensing
Warp was recently relicensed under the permissive Apache 2.0 license, making it more compatible with open-source projects like Slicer.
Complement to SOFA
The group discussed how SlicerWarp could serve as a lightweight and performant alternative or complement to SOFA, particularly since many of the integration patterns with Slicer are similar.