NVIDIA Launches Cosmos-H-Dreams: Real-Time Generative Simulator for Surgical Robots
NVIDIA
CMR Surgical
Cambridge Consultants
NVIDIA introduces Cosmos-H-Dreams, a real-time action-conditioned generative simulator for surgical robotics, built on the Cosmos-H-Surgical-Simulator foundation. It distills the capabilities into a causal few-step student model served via FlashDreams, enabling interactive surgical simulation on a single RTX PRO 6000 GPU at ~160 fps.
NVIDIA has released Cosmos-H-Dreams, a real-time generative simulator for surgical robotics. It distills the Cosmos-H-Surgical-Simulator world model into a causal student model that uses few-step diffusion, and serves it through the FlashDreams accelerated inference library. Running on a single NVIDIA RTX PRO 6000 GPU, it achieves interactive operation at approximately 160 frames per second. The simulator is action-conditioned, receiving an initial RGB frame and a live stream of robot kinematics to generate subsequent video frames autoregressively. It was demonstrated in collaboration with CMR Surgical and Cambridge Consultants on the Versius surgeon controller. The system supports human interfaces via browser (WebRTC) and Meta Quest (WebXR), and can also be connected to learned surgical policies. Cosmos-H-Dreams is designed to evaluate surgical policies, generate synthetic data, and enable closed-loop control for physical AI in surgery.
- Сокращения
- GPU = Graphics Processing Unit — графический процессор
- RGB = Red Green Blue — цветовая модель RGB
- dVRK = da Vinci Research Kit — исследовательский набор da Vinci
- JHU = Johns Hopkins University — Университет Джонса Хопкинса
- KV = Key-Value — ключ-значение
- WebRTC = Web Real-Time Communication — веб-коммуникация в реальном времени
- WebXR = Web Extended Reality — веб-расширенная реальность
- fps = frames per second — кадров в секунду
Source: Hugging Face blog —
original
