[International Developments] Medical robots enter the "Physical AI" era: NVIDIA GTC 2026 releases dedicated datasets and development blueprints

(Source: Robot Global News)

Medical robots enter the “Physical AI” era: NVIDIA GTC 2026 releases dedicated datasets and development blueprints

At the GTC conference held in March 2026, NVIDIA founder and CEO Jensen Huang announced the launch of a series of dedicated open platforms for medical robots. By releasing new physical AI datasets, foundational models, and development blueprints, NVIDIA aims to break down the technical barriers for medical robots from laboratory prototypes to clinical applications in operating rooms.

“Project Rheo”: The comprehensive development pathway for medical robots

The core highlight of NVIDIA’s release is a development blueprint called Rheo. It is an important part of the Isaac for Healthcare framework, designed to help developers build high-precision “digital twin hospitals.”

• Digital hospital simulation: Developers can quickly create physically accurate surgical room (OR) simulation environments through Rheo.

• Synthetic data generation: Using the Cosmos world model, Rheo can convert a small number of expert surgical demonstration videos into massive training datasets covering various extreme or rare clinical cases (Edge Cases).

• End-to-end validation: Robot strategies can undergo tens of thousands of virtual surgeries in simulation before deployment on physical hardware, greatly enhancing safety.

Core components: Open-H dataset and Cosmos-H model

To address the shortage of training data for medical AI, NVIDIA has partnered with several top medical institutions to launch:

Open-H dataset: Claimed to be the world’s largest dedicated dataset for medical robots, containing thousands of hours of surgical trajectories, detailed instrument operation data, and multimodal perception data.

Cosmos-H foundational model: Based on the Cosmos world model architecture, optimized specifically for operating room environments, capable of understanding complex physical laws (such as soft tissue deformation, fluid interactions, etc.).

GR00T-H visual-language action model: Provides the “brain” for medical robots, enabling them to understand doctor’s verbal commands and perform precise obstacle avoidance and coordination actions.

Industry alliances: Leading surgical giants gather

Currently, many top medical technology companies worldwide have announced integration into NVIDIA’s medical physical AI ecosystem:

• Traditional giants: Johnson & Johnson MedTech and Medtronic are utilizing this platform to optimize the dexterity of next-generation surgical robots.

• Emerging forces: CMR Surgical, Moon Surgical, and LEM Surgical (which develops “surgical humanoid robots”) are using these models to train their robot’s dual-arm coordination capabilities.

Industry perspective: From “assistive tools” to “intelligent co-pilots”

Huang emphasized in his speech: “Physical AI has officially arrived. In the medical field, this means robots will evolve from simple remote operation tools to ‘intelligent co-pilots’ with autonomous perception and collaboration capabilities.”

By opening these foundational datasets and models, NVIDIA not only consolidates its position as an AI computing infrastructure provider but also promotes the medical industry toward an era of “software-defined, AI-driven” precision medicine.

This report is based on official information released at the NVIDIA GTC conference in March 2026.

Massive information, precise insights, all on Sina Finance APP