NVIDIA's Medical Physics Simulation framework, part of Isaac for Healthcare, addresses three core challenges in healthcare robotics: data scarcity, generalization, and slow development cycles. The open-source, GPU-native framework offers two complementary simulation approaches. Classical solvers include an Endoluminal Simulation Module for catheter navigation using Cosserat rod mechanics and XPBD on the GPU, and a Surgical Simulation Module for soft-tissue deformation in procedures like cholecystectomy. Generative simulation via the Cosmos-H model family uses video diffusion transformers to produce realistic surgical video conditioned on robot kinematics, enabling real-time interactive environments at 30+ FPS. Both approaches integrate with NVIDIA Isaac Lab for large-scale RL policy training, supporting up to 512 parallel environments. Resources and GitHub repositories are provided to get started with catheter navigation workflows and Cosmos-H Dreams generative simulation.
