NVIDIA Accelerates Humanoid Robotics With Cloud-To-Robot Computing Platforms For Physical AI
In Brief
Nvidia CEO Jensen Huang announced Isaac GR00T N1.5 and the GR00T-Dreams blueprint to accelerate humanoid robot development.
At the COMPUTEX 2025 conference, Nvidia CEO Jensen Huang announced updates to the company’s foundational robotics technologies. Among these are Isaac GR00T N1.5, an enhanced version of Nvidia’s open and customizable foundation model for humanoid reasoning and task execution, and GR00T-Dreams, a blueprint designed to generate synthetic motion data to support physical AI training. These updates are part of a broader effort that includes Nvidia’s Blackwell systems, aimed at accelerating the development of humanoid robots.
The Isaac GR00T-Dreams blueprint enables the generation of synthetic motion sequences—referred to as neural trajectories—which physical AI developers can use to train robots in adaptable behaviors. Developers can refine a Cosmos Predict world foundation model (WFM) for their specific robot, and from a single input image, GR00T-Dreams creates videos showing the robot executing tasks in various environments. These simulations are then translated into action tokens, compact data segments that guide robotic learning.
This new blueprint builds upon the GR00T-Mimic blueprint introduced at the NVIDIA GTC conference in March. While GR00T-Mimic focuses on augmenting existing motion data using platforms like NVIDIA Omniverse and NVIDIA Cosmos, GR00T-Dreams centers on producing original motion datasets entirely within the Cosmos platform.
Isaac GR00T N1.5 represents the initial update to Nvidia’s adaptable and broadly applicable foundation model designed to support humanoid cognitive functions and task performance.
“Human demonstrations aren’t scalable — they’re limited by the number of hours in a day,” said Jensen Huang.
The GR00T-Dreams blueprint introduced a method for generating large volumes of synthetic motion data from single images. This approach supports more efficient robot behavior training by producing condensed data units known as action tokens.
The synthetic data generated through this method reduced the development time of the GR00T N1.5 model, which was completed in 36 hours—a process that would otherwise have taken close to three months.
The updated model demonstrates improved performance in executing standard tasks related to material handling and manufacturing and is expected to be compatible with Jetson Thor, scheduled for release later in the year.
New Robot Simulation And Data Generation Frameworks To Accelerate Training Pipelines
Additionally, NVIDIA has introduced a range of simulation technologies aimed at improving data accessibility and testing capabilities for physical AI systems.
Among these tools is NVIDIA Cosmos Reason, a world foundation model designed to apply chain-of-thought reasoning for generating accurate synthetic data, now available via Hugging Face. Furthermore, Cosmos Predict 2, an upgraded model used in the GR00T-Dreams blueprint, will soon be released with improved performance in world generation and reduced errors. NVIDIA has also introduced Isaac GR00T-Mimic, a system capable of producing large volumes of synthetic motion data from a limited number of human demonstrations.
An open-source dataset now offers 24,000 high-quality motion sequences that supported the training of GR00T N models. Isaac Sim 5.0, a simulation and data generation tool, is expected to become publicly accessible on GitHub, while Isaac Lab 2.2, an open-source robot learning environment, will include new tools for testing GR00T N models.
Disclaimer
In line with the Trust Project guidelines, please note that the information provided on this page is not intended to be and should not be interpreted as legal, tax, investment, financial, or any other form of advice. It is important to only invest what you can afford to lose and to seek independent financial advice if you have any doubts. For further information, we suggest referring to the terms and conditions as well as the help and support pages provided by the issuer or advertiser. MetaversePost is committed to accurate, unbiased reporting, but market conditions are subject to change without notice.
About The Author
Alisa, a dedicated journalist at the MPost, specializes in crypto, AI, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.
More articles
Alisa, a dedicated journalist at the MPost, specializes in crypto, AI, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.
