Nvidia's open-source simulator reduces the training time for robots to just 2 minutes.

      The challenging aspect of medical robotics lies not in the robot itself, but in the practical application. A surgical system must undergo thousands of trials to master a delicate procedure, and these cannot be conducted on actual patients. Nvidia believes the solution is to enable the robot to practice within a simulated body, millions of times.

      The company has introduced an open-source Medical Physics Simulation framework, which is part of its Isaac for Healthcare platform, according to HIT Consultant. This framework simulates the interactions between instruments and anatomy, allowing developers to train and test physical-AI policies without engaging with hardware.

      The speed is noteworthy. By utilizing 8,192 training environments simultaneously on GPUs, Nvidia claims to achieve a remarkable acceleration. This reduces the time needed for robotic policy training from over five hours to less than two minutes, marking a shift from an overnight task to a quick coffee break.

      However, the more significant aspect is what the simulation yields: evidence. As the framework is open, developers have access to the data, models, and weights they require. They can leverage these resources to establish the verifiable proof that medical authorities expect.

      The framework integrates two types of physics. Classical physics, through Nvidia’s Warp and Newton engines, addresses the hard mechanics, including friction, contact, and the resistance of tissues against a catheter or guide wire.

      On the other hand, generative AI physics, via a model named Cosmos-H Dreams, forecasts the more complex scenarios, such as the deformation of soft tissue. It can even incorporate simulated fluoroscopy, X-ray, and ultrasound.

      Prominent users add credibility to the framework. Surgical robotics companies like Medtronic, Johnson & Johnson MedTech, and CMR Surgical are adopting it for surgical digital twins and endovascular training. Chris Fryer, CTO of CMR Surgical, remarked that open-source models enable companies to "build on shared knowledge, accelerating responsible innovation."

      There’s a strategic reasoning behind this generosity. By open-sourcing the tools, Nvidia establishes itself as the foundational infrastructure for the upcoming generation of surgical robots, all of which will operate on Nvidia GPUs.

      Nonetheless, a simulation is not a substitute for a patient. A policy that excels in a digital twin must still prove its effectiveness in an operating room, and regulators have yet to clarify how much synthetic evidence they will deem acceptable.

      Despite this, the trajectory is noteworthy. The main obstacle in medical AI has been data, rather than concepts, and Nvidia has provided the industry with a free path around it. If regulators determine that a robot trained in a virtual environment can safely function in a real one, the two-minute rehearsal could revolutionize how surgical machines are developed.