NVIDIA: Open-Source Medical Physics Framework Cuts Robot Policy Training from 5 Hours to 2 Minutes

What does the new open-source medical physics framework bring?

NVIDIA has open-sourced the Medical Physics Simulation framework within its NVIDIA Isaac for Healthcare platform. The framework combines classical physics simulation, mathematical modeling of real-world forces and materials, with the generative AI system Cosmos-H Dreams, which generates realistic visual and sensor scenarios. The goal is to train robot policies, learned decision-making models for robotic actions, before costly and slow testing on real surgical hardware.

Benchmark cuts training from 5 hours to 2 minutes

According to NVIDIA’s benchmark, running 8,192 parallel training environments simultaneously cut training time from more than 5 hours to under 2 minutes. The comparison is direct: the same task that previously required sequential testing over hours of work is now solved through parallel simulation within a single short run, accelerating iteration in the development of robotic algorithms for surgical applications by orders of magnitude.

Partners in surgical robotics and the technologies behind it

Partners using the framework include CMR Surgical and Cambridge Consultants, who contributed nearly 500 hours of anonymized clinical data for soft-tissue surgical simulation, and Johnson & Johnson MedTech, whose MONARCH system for urology uses a digital twin, a virtual replica of the physical system, for testing before deployment. Additional partners include XCath, Inner Logic, and Medtronic Structural Heart, each with its own specific robotic platforms for minimally invasive procedures. The framework is built on NVIDIA’s CUDA, Warp, Newton, and Cosmos technologies, linking it to NVIDIA’s broader infrastructure for robotics and simulation. Open-sourcing the code means development teams worldwide can build their own medical robotic simulations without needing their own costly infrastructure for capturing real clinical data, speeding up the entire development cycle from idea to clinically validated robotic system.