NVIDIA's High-Fidelity Physics Research<https://research.nvidia.com/labs/prl/> group works at the intersection of physics simulation, computer graphics, and machine learning. We prioritise accuracy and fidelity over real-time performance, and we bridge the gap with sparse and neural representations and other ML techniques. We study both forward and inverse physics problems, including fluids, cloth, hair, soft bodies, oceans, and granular materials such as sand. We are also interested in bringing more accurate physics into neural representations of the world. The group is led by Ken Museth, creator of OpenVDB, and includes researchers and engineers across Asia Pacific, Europe, and North America. Our work includes the open-source OpenVDB, NanoVDB and fVDB libraries, which are widely used in film VFX and scientific computing. (Our publications<https://research.nvidia.com/labs/prl/publication/>.)

Internship projects span research and engineering. Broad priority areas this cycle (I am most involved with the first 3 below) include:

  *   Agentic workflows: Evaluation approaches for AI agents on long, complex tasks, and tools that turn research papers into working code. Measuring the impact on software / skill changes on agent performance in complex 3D reconstruction and simulation tasks.

  *
Geometry processing and sparse data structures: GPU-accelerated infrastructure around NanoVDB and VDB. State-of-the-art isosurfacing. Using tensor cores for PDE and finite-element operators on sparse grids.
  *   3D learning and real-to-sim: Learning on sparse volumetric data. Making reconstructed scenes and assets ready for physics simulation, for example through segmentation and component libraries.

  *   Physics simulation: Adaptive-resolution solvers (MPM and lattice Boltzmann) for granular and multiphase flows. Large-scale ocean and hurricane simulation. Reduced-order methods for cutting and fracture.

  *   Robotics and simulation: Kinematics and constraint models for robot simulation. Estimating material parameters, such as friction and stiffness, from haptic sensing.

  *   Monte Carlo methods for scientific simulation: Grid-free PDE solvers such as Walk on Spheres, with applications such as semiconductor and electronic-design simulation.

I would be grateful if you could pass this on to students who may be a good fit or recommend students directly. Strong candidates are typically PhD or MSc (or highly skilled final-year undergraduate) students with experience in graphics, simulation, numerical methods, GPU programming, or ML. Paid internships typically cover the US summer season May-August but there is potential for flexibility for international students. Internships may be US or Australia (WFH or University) based, depending on visa situation. To apply or be considered, students can contact me (Mark Harris) directly at mharris@nvidia.com by November 20.