Models & Policies

Pre-trained policies and evaluation frameworks for healthcare robotics applications

The following pre-trained models are used in our End-to-End Workflows and are hosted on Hugging Face by NVIDIA.

GR00T-H

NVIDIA GR00T foundation model post-trained for healthcare robotics — the base architecture behind the fine-tuned policies below.

GR00T-H GR00T N1.6 post-trained on Open-H-Embodiment for surgical robotics — 3B params, 7 embodiments, 601h training data

Robotic Ultrasound Policies

VLA models fine-tuned for autonomous ultrasound scanning in simulation.

GR00T post-trained for liver scan GR00T-N1 VLA with Cosmos data augmentation — 2.2B parameters

Pi0 post-trained for liver scan π0 VLA with Cosmos-transfer augmented data — 3B parameters

Hospital Automation Robotic Policies

VLA models for hospital automation tasks including surgical instrument handling, cart transport, and trocar assembly.

GR00T-N1.6-Rheo Pick-N-Place Tray Pick-and-place of sterilized box from shelf to cart — 3B params, G1 embodiment

GR00T-N1.6-Rheo Sim Push Cart Cart transport with handle grasping and navigation — 3B params, G1 embodiment

GR00T-N1.5-RL-Rheo Assemble Trocar Trocar assembly from surgical tray to Mayo Stand — 3B params, multi-camera

SO-ARM GR00T Surgical instrument handling with SO-ARM101 — 3B params, sim + real training

---

Model retrieval is typically done automatically when running each workflow. You can also download models manually (e.g. huggingface-cli download <repo_id>).