NVIDIA open-sources first GPU-accelerated framework for medical physics simulation

NVIDIA has announced the release of Medical Physics Simulation, a new open-source framework within NVIDIA Isaac for Healthcare that leverages GPUs to accelerate medical physics simulation for the first time. The platform enables modeling of interactions between anatomy and medical devices, including contact, friction, and sensory data, as well as generating hard-to-reproduce scenarios for training and evaluating robot behavior. The framework combines classical physics simulation and generative physics based on NVIDIA Cosmos-H Dreams, allowing the creation of realistic virtual environments without needing to rebuild them for each workflow. The open-source code provides transparency for regulatory bodies and reproducibility of results. Tests show that 8192 parallel training environments on a GPU reduce training time from five hours to under two minutes. Developers are already using the framework: CMR Surgical and Cambridge Consultants use Cosmos-H-Dreams to train soft-tissue surgical procedures, Johnson & Johnson MedTech builds digital twins of the MONARCH platform, and XCath trains policies for endovascular autonomy. The vendor also highlights CMR Surgical's contribution to the Open-H Embodiment open dataset. The framework can be used independently or alongside digital twins, medical sensor simulation, and the Isaac Lab library.