Nvidia launches open-source simulator for healthcare robots

Nvidia added an open-source Medical Physics Simulation framework to its Isaac for Healthcare platform to generate the physical interactions needed to train surgical and diagnostic robots. The framework combines classical physics simulation with a generative component called Cosmos-H Dreams and runs on Nvidia’s GPU libraries Warp and Newton for parallel execution.

Classical simulation models mechanical behaviour such as catheter bending and tissue contact forces. The generative component reproduces visual scene dynamics and anatomical variation that are hard to encode by hand. Nvidia reported a benchmark in which 8,192 parallel environments reduced training time from more than five hours to under two minutes.

Nvidia describes the framework as a way to produce simulated failure modes and rare scenarios on demand, including a guidewire catching on a calcified vessel wall, a kidney stone lodged at an unusual angle, and uncommon soft-tissue responses that appear in a small share of procedures.

Early users apply the framework in different ways. CMR Surgical and Cambridge Consultants contributed close to 500 hours of anonymized clinical data from CMR’s Versius Surgical Robotic System to the Open-H Embodiment dataset covering cholecystectomy, prostatectomy, hernia repair and hysterectomy, and are using Cosmos-H Dreams to model soft-tissue interactions and patient-specific simulations. Johnson & Johnson MedTech is building a digital twin of its MONARCH endoluminal platform focused on kidney-stone scenarios. XCath is training endovascular autonomy policies for navigating blood vessels. Inner Logic is generating synthetic data to validate device mechanics and says it intends to produce in silico evidence to support regulatory filings, though no such submission has been confirmed publicly. Medtronic Structural Heart has explored simulated X-ray sensing for catheter navigation research. None of the organisations listed has reported deploying a system trained with this framework on patients, and Nvidia has not claimed clinical deployment.

Developers and some users say publishing the code helps regulators and review boards inspect simulation logic, reproduce results and build an evidence trail for regulatory review. Chris Fryer, CTO at CMR Surgical, described the value of open-source development: “Open-source models allow us to build on shared knowledge, accelerating responsible innovation, and gives us the potential to deliver more consistent care and better outcomes for patients worldwide.”

Researchers and companies caution that high throughput does not prove clinical reliability. A policy that performs well in parallel simulation may still fail in clinical settings where imaging can be incomplete, sensors delayed, or anatomy differs from the models. Validation that simulated failure modes match real-world failures requires physical testing and clinical evidence, and the companies involved have not published such validation data.

Nvidia says the framework can shorten the pre-hardware phase of development by running many parallel training environments rather than rebuilding a custom simulation scene for each workflow. The company presents the work as an approach for giving robotic systems experience of contact, force and consequence without exposure to live procedures during early development.