NVIDIA Introduces Cosmos3-H-Dreams for Interactive Generative Simulation of Clinical Surgery
NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative Simulation
- Who
- NVIDIA
- What
- introduces Cosmos3-H-Dreams, an interactive generative simulator for clinical cholecystectomy
What happened
NVIDIA introduces Cosmos3-H-Dreams, a surgical simulation model trained on clinical cholecystectomy video and robot kinematics from the CMR Surgical Versius platform. Powered by Cosmos 3 and accelerated with FlashDreams, it supports live instrument control on a single NVIDIA RTX PRO 6000 GPU, with separately trained four-step and two-step checkpoints balancing visual quality and inference speed. It is a research simulator, not a validated system for patient care.
Why it matters
The separately trained two-step and four-step checkpoints expose a concrete tradeoff between inference speed and visual quality for researchers adapting interactive surgical simulation to their own data.
The story so far
NVIDIA introduced Cosmos3-H-Dreams, a model that supports live control of simulated surgical instruments on a single NVIDIA RTX PRO 6000 GPU. It is a research simulator for cholecystectomy, or gallbladder removal, and is not validated for patient care. The model was trained on clinical cholecystectomy video and robot kinematics from CMR Surgical’s Versius platform. Built on Cosmos 3 and accelerated with FlashDreams, it offers separately trained four-step and two-step checkpoints that balance visual quality and inference speed. NVIDIA also provides a guide for training a real-time student model on a user’s own dataset through teacher fine-tuning and Self-Forcing distillation.
Keep reading
Latest daily · Issue 27TorchNeuron brings native PyTorch support for Hugging Face models on AWS TrainiumTrending now
Tomorrow at 08:00, a briefing like this arrives in your inbox.
The same kind of briefing as the one above: a few stories worth reading. Free. Mail starts after you confirm, and every letter has a one-click unsubscribe. Read a past issue · 2026-10-20
