Nvidia advances with Humanoid Robotics with the help of the cloud

Nafidia He said that he is advancing with Human robots Technology, providing a basic model dedicated to human thinking, a plan to generate artificial movement data, and more Blackwell systems to accelerate the development of human robot.

At the computex 2025 commercial exhibition in Taiwan, NVIDIA unveiled ISAAC GR00T N1.5, which is the first update of the open NVIDIA model, generalized and fully customized for thinking and human skills; Nvidia ISAAC GR00T-DREAMS, which is a plan to create artificial movement data; And Nvidia Blackwell Systems to accelerate the development of human robot.

Humanoid and Robotics Agility Robotics ,, Boston Dynamics, Fourrier, Foxlink, Galbot, Mentee Robotics, Neura Robotics, Robotics General, Skild Ai and Xpeng Robotics adopt NVIDIA technologies Isaac Platform to enhance the development and development of human robot.

“Amnesty International and material robots will achieve the following industrial revolution,” Nvidia, CEO of NVIDIA, said in a statement. “From the brains of artificial intelligence of robots to the simulation worlds to practice in or Amnesty International to reach the forms of the Training Corporation, NVIDIA provides building blocks for each stage of the stages of the development of robotics.”

The new ISAAC GR00T data is concluded the data gap

NVIDIA ISAAC GR00T-DREAMS dreams that were shown at the main computex in Huang, a scheme that helps generate huge amounts of artificial movement data-also known as nervous paths-Artificial intelligence developers can use to teach robotics new behaviors, including how to adapt to variable environments.

The developers can first predict the POMS models (WFMS) after their robot. Next, using one image when input, the GR00T-Driends creates videos of the robot that performs new tasks in new environments. The scheme then extracts the distinctive symbols – parts of compressed data, which are used – which are used to teach robots how to perform these new tasks.

The GR00T-DREAMS chart complements the ISAAC GR00T MIMIC chart, which was released at the NVIDIA GTC conference in March. While GR00T MIMIC uses NVIDIA OMNINIVERSE and Nvidia Cosmos to increase current data, GR00T dreams use Cosmos to create completely new data.

“NVIDIA has a very strong mechanical strategy, and it is centered on what Jensen calls the three computer problem,” Jim van, director of Amnesty International and the distinguished scientist at NVIDIA.

He pointed out that the company has an OVX computer, which aims to conduct a simulator and graphic simulation engine, and is used to collect and create data. This data is consumed by a DGX computer, which is used to train the foundation forms. Then it is posted on the HX computer, which is the operation time on the edge of platforms like Humanoid Robots.

Van Gr00T is an Amnesty International and robot -based work course.

“It is an equal to the three computing problem,” he said.

He pointed to two progress in Project Gr00T, GR00T DREAMS and GR00T N1.5. (He said he is very proud of these names).

For GR00T Dreams, Fan said it is a model that can create videos for robots training. A collection of videos has been shown and said that all videos were created by Nvidia Cosmos.

) And now this becomes artificial data to increase our real robot data groups. Many of you may know, the collection of data on the real robot takes a long and expensive time, because you are 24 hours limited per day per day, right? It is a basic material system in this limit of physical restrictions.

The result is that the robots will be able to capture objects properly. You can tell her to capture an option, pour some orange juice, or open a laptop. Van said that the robot was not trained in these specific procedures, but because he was trained in video models, the robot is able to “understand physics and the meaning of these actions.”

Thus he learns how to implement procedures.

New Isaac GR00T models offer the development of human robot

NVIDIA Research used the GR00T-Dreams chart to create artificial training data to develop GR00T N1.5-update by GR00T N1- in just 36 hours, compared to what would take nearly three months without the plan.

GR00T N1.5 can better adapt to new environments and work space, as well as identify objects through user instructions. This update greatly improves the model success rate of common material processing tasks and manufacturing tasks such as sorting or setting objects. GR00T N1.5 can be published on Jetson Thor, as it was later launched this year.

The GR00T N1.5 foundation combines GR00T dreams as part of the artificial data generation pipeline. Van said that NVIDIA has upgraded the backbone of the visual language, so GR00T N1.5 will have a better ability to adapt and comply with language teaching.

GR00T N1.5 will be announced in Computex and then released open source by June 9. As for GR00T Dreams, NVIDIA is still working on the schedule. Van said that hope is to open the source as much as possible.

It includes the first adoption of GR00T N Aeirobot, Foxlink, Lightwheel and Neura Robotics. AEirobot emphasizes the ALIICE4 model to understand natural language instructions and implement complex work for selection and place in industrial settings. The Foxlink group uses to improve the ability of the industrial robot processor and Yiuki, while Lightwheel is harnessing it to verify the authenticity of artificial data to spread the fastest human robot in factories. Robotics Neura evaluates the model to accelerate the development of home automation.

New robot simulation and data generation

The development of high skills human robots requires a huge amount of diverse data, which is costly to capture and process. Robots should be tested in the material world, which can provide costs and risks.

To help bridge the gap and test, NVIDIA revealed the following simulation techniques:
● The cause of Nvidia Cosmos, a new WFM that uses a series of logical ideas to help organize accurate and high -quality artificial data to train in the artificial intelligence model
● Cosmos Problect 2, used in GR00T dreams, will also come to face embrace with high -quality global performance improvements and reduce hallucinations.
● Nvidia ISAAC GR00T MIMIC, which is a plan to generate large quantities of artificial movement paths to process robot, using a few human demonstrations.
● AI’s open source material data collection, which now includes 24,000 high -quality human robot movement used to develop GR00T N.
● Nvidia ISAAC SIM 5.0, Simulation and Generational Data generation frame, now publicly available on GitHub.
● Nvidia ISAAC LAB 2.2, which is an open source robot learning framework, which will include new evaluation environments to help developers to test the GR00T N.

Foxconn and FoxLink uses the GR00T-Mimic scheme to generate artificial movement to accelerate robots training pipelines. Agility Robotics, Boston Dynamics, Fourrier, Mentee Robotics, Neura Robotics, Xpeng Robotics simulates and training human robots using Nvidia ISAAC SIM and ISAAC LAB. Skild AI uses simulation frameworks to develop a general robot
Intelligence, general robots merge them on the robot intelligence platform.

Blackwell Global Robot developers

World systems manufacturers build work stations, work servers and NVIDIA RTX Pro 6000, which leads to the operation of one structure to operate every burden of robot development easily through training, generating artificial data and learning robot and simulation.

Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo, SuperMicro have announced the Nvidia RTX Pro 6000 Blackwele, Dell Technologies and Lenovo Services. Blackweell.

When more account is needed to run widespread work burdens or data generation, developers can take advantage of NVIDIA Blackwell systems such as GB200 NVL72- available with NVIDIA DGX Cloud on leading cloud services and NVIDIA Cloud-to achieve up to 18X Data processing performance.