nvidia just turned a 5-hour robot training run into under 2 minutes — for surgical robots

A new open-source physics engine lets Nvidia GPUs simulate thousands of surgical robot training runs in parallel — and hospitals, not gamers, are… Aliteq Lena Fischer · AI & Local Compute Editor

Nvidia's Medical Physics Simulation framework ran 8,192 parallel robot-training environments and cut that run from over five hours to under two minutes.

The numbers that matter

Single-environment physics runs at roughly 1,300 Hz; at 512 parallel environments it holds 60 Hz — fast enough for real-time reinforcement learning.

The numbers that matter

It's open source, built into Nvidia Isaac for Healthcare, and pairs classical physics (Warp, Newton) with a generative video model called Cosmos-H for realistic visual simulation.

The numbers that matter

Johnson & Johnson MedTech, CMR Surgical, Medtronic Structural Heart, XCath and Inner Logic are already building on it.

The numbers that matter

Nvidia frames it as an answer to medical robotics' four-to-seven-year development cycle, most of which currently goes to acquiring and validating real surgical data.

The numbers that matter

This is a more interesting use of Nvidia's compute advantage than another gaming GPU price hike, and I mean that as someone who covers the price hikes constantly. The company's moat is CUDA plus raw…

Read the full story

Aliteq nvidia just turned a 5-hour robot training run into under 2 minutes — for surgical robots Read the full story on Aliteq