NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative Simulation
Our first Cosmos-H-Dreams release, built on Cosmos-Predict2.5, brought interactive generative simulation to dVRK tabletop suturing, a foundational proving ground for surgical robotics.
NVIDIAHugging FaceKnowledge distillationnvidia/Cosmos-H-Dreamsnvidia/Cosmos3-H-Embodied-Nanonvidia/Cosmos3-H-Dreams
ProofWeights ↗
Key points
- As world foundation model (WFM), Cosmos-H-Dreams demonstrated an alternative to numerical simulation: rather than explicitly modeling scene geometry, tissue deformation, and physical interactions, it learned visual dynamics directly from synchronized video and robot kinematics, yielding interactive tabletop task simulation with high realism.
- Today, Cosmos3-H-Dreams advances this capability by learning from video and robot kinematics recording during clinical procedures on the CMR Surgical Versius platform.
- Cosmos3-H-Dreams builds on Cosmos3-H-Embodied-Nano, a 16-billion-parameter model pretrained on Open-H-Embodiment, and specializes a bidirectional teacher on curated cholecystectomy video and kinematics.
- nvidia published the model Cosmos3-H-Embodied-Nano on Hugging Face.
Sources (3)
- [1]NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative SimulationHugging Face Blog · Oct 20, 06:17 PM
Our first Cosmos-H-Dreams release, built on Cosmos-Predict2.5, brought interactive generative simulation to dVRK tabletop suturing, a foundational proving ground for surgical robotics.
As world foundation model (WFM), Cosmos-H-Dreams demonstrated an alternative to numerical simulation: rather than explicitly modeling scene geometry, tissue deformation, and physical interactions, it learned visual dynamics directly from synchronized video and robot kinematics, yielding interactive tabletop task simulation with high realism.
- [2]nvidia/Cosmos3-H-Embodied-NanoHF: NVIDIA · Oct 19, 12:45 PM
nvidia published the model Cosmos3-H-Embodied-Nano on Hugging Face.
- [3]nvidia/Cosmos3-H-DreamsHF: NVIDIA · Oct 19, 12:35 PM
Extractive summary: sentences quoted from the sources.
Hype check
Questioned1 reported issue since
- Issue · Oct 19, 2026A hacker broke into John Deere and SharkNinja with an army of AI agents
- Evaluation · Oct 19, 2026Selective Transfer through Sequence Distillation: From Benchmark Knowledge to Backdoor Behaviour
- Evaluation · Oct 19, 2026The Model Swap Workshop: GPT-5.5 vs Kimi vs DeepSeek vs Mistral
- Evaluation · Oct 19, 2026Beyond Behavioral Accuracy: Evaluating Individual-Level Fidelity in LLM Investor Simulation
- Evaluation · Oct 19, 2026Agentic Orchestration of EDC Data Space Retrieval and Natural Language Driven Data Processing
- Evaluation · Oct 19, 2026WhatIf-Bench: Benchmarking Intervention-Aware Time Series Forecasting
Before this
- Oct 19, 2026nvidia/cosmos3-nano-libero10-sdg-export-S5_2_s43
- Oct 19, 2026TorchNeuron: running HF models on AWS Trainium is easier than ever
- Oct 19, 2026From Short Clips to Real-Time Streams: Identity-Aware Supervision and Context-Efficient Generation for Video Face Swapping
- Oct 18, 2026GPT-2 KV cache from scratch: correctness, latency, and memory measurements [P]
- Oct 17, 2026Nemotron 3.5 Bolt, Muse Spark Open, and a Fast LFM2.5 24B
- Oct 16, 2026nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-MXFP8
