NVIDIA has unveiled a powerful suite of cloud-to-robot platforms designed to accelerate the development of humanoid robots, as part of its continued push into Physical AI. At the heart of the announcement is the updated Isaac GR00T N1.5 model, new synthetic data tools like GR00T-Dreams, and universal Blackwell-powered RTX PRO 6000 systems that streamline simulation, training, and deployment.
NVIDIA Empowers the Next Wave of Physical AI
A growing roster of robotics leaders — including Agility Robotics, Boston Dynamics, Foxconn, NEURA Robotics, and XPENG Robotics — are adopting NVIDIA’s Isaac™ platform technologies to develop next-gen humanoid robots.
“Physical AI is the foundation of the next industrial revolution,” said Jensen Huang, founder and CEO of NVIDIA. “Our end-to-end platforms—from AI brains to simulation and training supercomputers—provide the full toolset needed for the robotics era.”
GR00T-Dreams Blueprint Enables Fast Synthetic Training Data Generation
Introduced during Huang’s keynote, GR00T-Dreams is a synthetic data generation blueprint that uses Cosmos Predict world foundation models (WFMs) to create motion data — known as neural trajectories — from a single image input. The result is task-specific video content and behavior tokens, enabling robots to learn new tasks quickly and safely.
This complements the earlier GR00T-Mimic blueprint, which enhances training using a few real human demonstrations combined with synthetic augmentation in NVIDIA Omniverse™.
GR00T N1.5: Next-Level Humanoid Reasoning
Fueled by data from GR00T-Dreams, NVIDIA trained GR00T N1.5 in just 36 hours — a process that would typically take months with manual data collection. The model shows major improvements in understanding instructions, object recognition, and material handling tasks, boosting effectiveness in environments like warehouses or factories.
Companies like Foxlink, Lightwheel, and NEURA Robotics are integrating GR00T models to power real-world robots with natural language processing and enhanced dexterity.
Accelerated Simulation and Testing for Real-World Readiness
Recognizing the cost and risk of real-world testing, NVIDIA introduced key simulation updates:
- Cosmos Reason: A WFM for better synthetic data curation, available now on Hugging Face.
- Cosmos Predict 2: An upgraded world model for environment generation, coming soon.
- Isaac GR00T-Mimic: Generates large-scale motion data from minimal real demos.
- Isaac Sim 5.0: Soon open on GitHub, this is a robust tool for robot simulation and synthetic data.
- Isaac Lab 2.2: Adds test environments for validating GR00T N models.
RTX PRO and Blackwell Systems Fuel the Robotics Pipeline
Robot developers can now use NVIDIA RTX PRO 6000 Blackwell-powered workstations and servers, announced by major OEMs including Dell, Lenovo, Cisco, and Supermicro, to power every stage of robot development — from training and simulation to deployment.
When more compute is needed, developers can scale to NVIDIA GB200 NVL72 systems via DGX Cloud, delivering up to 18x performance gains in large-scale data processing.
For real-time operation, NVIDIA’s Jetson Thor platform will enable on-device inference using trained foundation models.
Learn More:
Watch the COMPUTEX 2025 keynote or visit NVIDIA’s GTC Taipei coverage for full updates on the Isaac robotics platform.