The First Healthcare Robotics Dataset and Foundational Physical AI Models for Healthcare Robotics

Traditional perception-based AI models are insufficient for healthcare robotics, which requires data reflecting physical interaction and embodiment. To address this, the Open-H-Embodiment dataset has been created as a large-scale, open resource for training physical AI in surgical and ultrasound robotics. Built on this data, GR00T-H is a new vision-language-action model capable of performing complex surgical tasks like suturing. Additionally, the Cosmos-H-Surgical-Simulator, a world foundation model, generates physically plausible surgical video from actions to bridge the sim-to-real gap and create synthetic training data. These initiatives aim to provide the foundational tools necessary to advance the field of physical AI in healthcare.