Nvidia opens medical physics simulation framework for robots

Nvidia has released its Medical Physics Simulation framework as open source within Isaac for Healthcare. The software is aimed at developers building medical robots.

The framework is designed to let teams model interactions between anatomy and medical devices, simulate imaging and sensor inputs, and train or test robot behaviour before moving to hardware. It brings together physics simulation, robot learning and generative AI methods in a single environment.

Medical robotics developers face a persistent data problem: rare clinical scenarios are difficult to capture, and physical testing is expensive and time-consuming. Robots intended for healthcare settings must also cope with wide variation in anatomy, changing conditions during procedures and the imperfect imaging that often guides interventions.

Nvidia is positioning the software as reusable infrastructure rather than a set of custom-built simulation scenes for each workflow. Because the framework is open source, developers can inspect, adapt and extend the code for their own devices and clinical applications.

Training scale

The software runs on Nvidia GPUs and sits within the broader Isaac for Healthcare stack. Benchmark results showed 8,192 robot training environments could run in parallel, reducing one training process from more than five hours to less than two minutes.

That scale matters for teams trying to identify failure modes earlier in development. Rather than relying only on limited physical experiments, developers can generate large numbers of difficult or uncommon scenarios in software and assess how robot policies respond.

The framework is built to support modelling of anatomy, device contact, friction, motion and sensor behaviour. It can also connect vascular anatomy, flexible instruments such as catheters and guidewires, simulated X-ray imaging and reinforcement learning, while remaining open to other devices, anatomies and sensing methods.

Mixed approach

A central feature of the release is its combination of conventional physics simulation with generative AI-based physics simulation. Classical methods are intended to represent known physical rules, while Nvidia's Cosmos-H Dreams component is meant to model visual scene dynamics learned from procedural data in real time.

The approach reflects a wider shift in robotics towards training and evaluation in virtual environments before extensive laboratory work begins. In healthcare, that can be especially relevant because teams need evidence of how systems behave across different anatomies and edge cases, and because transparency around models and data remains important in regulatory review.

Several medical technology groups are already using simulation-led development built around parts of the stack. Among them, CMR Surgical and Cambridge Consultants, part of Capgemini, are using Cosmos-H-Dreams for soft-tissue surgical procedures and patient-specific simulations.

CMR has also contributed nearly 500 hours of anonymised clinical data from its Versius Surgical Robotic System to the Open-H Embodiment dataset. The contribution covers procedures including cholecystectomy, prostatectomy, hernia repair and hysterectomy.

"Open source models allow us to build on shared knowledge, accelerating responsible innovation and, ultimately, gives us the potential to deliver more consistent care and better outcomes for patients worldwide," said Chris Fryer, Chief Technology Officer, CMR Surgical.

Other groups named by Nvidia include Johnson & Johnson MedTech, which is using Medical Physics Simulation and a Cosmos-based foundation model to build digital twins of its MONARCH endoluminal platform for urology. This work includes modelling complex anatomy and kidney stone scenarios.

According to Nvidia, XCath is using the framework for endovascular autonomy policy training. Inner Logic is using it for synthetic data generation, device mechanics validation and the production of in silico evidence intended to support regulatory pathways.

Medtronic Structural Heart is also exploring use of the framework with simulated X-ray sensing to generate data for catheter navigation research. That points to one of the software's more immediate uses: building virtual test environments for image-guided procedures, where device motion, anatomy and sensor feedback interact in complex ways.

Broader stack

Medical Physics Simulation can be used on its own or alongside other parts of Isaac for Healthcare, including digital twin workflows, medical sensor simulation, Isaac Lab and Nvidia's open models and policies. By making the framework modular, Nvidia is trying to reduce the bespoke engineering typically needed to set up robotics simulation in healthcare.

The move also expands Nvidia's presence in medical robotics, an area where developers increasingly need large volumes of training and evaluation data but face obvious limits on collecting that data from real procedures. The framework is intended to help them create and reuse simulation environments across devices, anatomies and healthcare robotics applications.