Nvidia charges ahead with humanoid robotics aided by the cloud

Nvidia is charging full speed ahead into the realm of humanoid robotics, and they're not holding back. At the Computex 2025 trade show in Taiwan, they unveiled a series of innovations that are set to redefine the landscape of robotics development. Among these, the star of the show was Nvidia Isaac GR00T N1.5, the latest iteration of Nvidia's open, fully customizable foundation model for humanoid reasoning and skills. Alongside it, Nvidia introduced Isaac GR00T-Dreams, a blueprint designed to generate synthetic motion data, and the powerful Nvidia Blackwell systems, which are poised to accelerate the development of humanoid robots.

A number of leading humanoid and robotics developers, including Agility Robotics, Boston Dynamics, Fourier, Foxlink, Galbot, Mentee Robotics, NEURA Robotics, General Robotics, Skild AI, and XPENG Robotics, are already on board, leveraging Nvidia's Isaac platform technologies to push the boundaries of what's possible with humanoid robots.

Jensen Huang, Nvidia's CEO, couldn't contain his excitement, stating, "Physical AI and robotics will bring about the next industrial revolution. From AI brains for robots to simulated worlds to practice in, or AI supercomputers for training foundation models, Nvidia provides building blocks for every stage of the robotics development journey."

New Isaac GR00T Data Generation Blueprint Closes the Data Gap

Ever wondered about having your own humanoid robot? During his keynote at Computex, Huang showcased Nvidia Isaac GR00T-Dreams, a game-changing blueprint that churns out vast quantities of synthetic motion data, or neural trajectories. This data is a goldmine for physical AI developers, helping them teach robots a whole range of new behaviors and how to adapt to ever-changing environments.

The process starts with developers post-training Cosmos Predict world foundation models (WFMs) for their robots. Then, with just a single image as the input, GR00T-Dreams generates videos of the robot tackling new tasks in new settings. It then extracts action tokens—bite-sized, easily digestible pieces of data—that the robots use to learn these new tasks.

GR00T-Dreams works hand-in-hand with the Isaac GR00T-Mimic blueprint, introduced earlier at the Nvidia GTC conference in March. While GR00T-Mimic leverages Nvidia Omniverse and Nvidia Cosmos to enhance existing data, GR00T-Dreams uses Cosmos to create entirely new data from scratch.

Jim Fan, Nvidia's director of AI and distinguished scientist, shared his enthusiasm in a press briefing, "Nvidia has a very strong robotic strategy, centered around what Jensen calls the three computer problem." He explained that the OVX computer handles simulation and graphic simulation physics engines, generating data that the DGX computer then uses to train foundation models. This data is then deployed to the HX computer, which runs the show on the edge for platforms like humanoid robots.

Fan proudly referred to Gr00t as the lifecycle of physical AI and robot-based workflows, emphasizing, "It is an instantiation of the three-computer problem." He highlighted two major advances in Project Gr00t: Gr00t Dreams and Gr00t N1.5, jokingly adding that he was quite proud of those names.

For Gr00t Dreams, Fan described it as a model that generates videos to train robots. He showcased numerous videos, all generated by Nvidia Cosmos, explaining, "We found a way to apply advanced video generation models like Cosmos to help humanoid robotics. So on a high level, how this method works is we first fine-tune Cosmos on robot videos from our lab so that this video model is now customized to the robots at our lab. Then we can use this fine-tuned model to generate, in principle, an infinite number of dream videos by prompting the model in different ways. And now that becomes synthetic data to augment our real robot data sets. As many of you might know, collecting data on the real robot is very time-consuming and costly because you're fundamentally limited by 24 hours per robot per day, right? It's a physical system, but with Gr00t Dreams, this new workflow, this new set of algorithms, now we're able to break this fundamental physical limit and then multiply data at an unprecedented scale next."

The result? Robots that can pick up objects correctly, whether it's a cucumber, pouring orange juice, or opening a laptop. These are actions the robot has never been trained on, Fan noted, but thanks to training with video models, the robot can "understand the physics and the meaning of these verbs" and learn to perform them.

New Isaac GR00T Models Advance Humanoid Robot Development

Nvidia's GR00T-Dreams blueprint isn't just talk; it's been put to work, generating synthetic training data to develop GR00T N1.5 in a mere 36 hours—a task that would have taken nearly three months without the blueprint. GR00T N1.5 can now better adapt to new environments and workspace configurations, as well as recognize objects through user instructions. This update significantly boosts the model's success rate for common tasks like sorting or putting away objects, and it's ready to be deployed on Jetson Thor, set to launch later this year.

The GR00T N1.5 foundation model integrates Gr00t Dreams into its synthetic data generation pipeline. Nvidia has upgraded the visual language backbone, ensuring that GR00T N1.5 will have superior adaptability and better compliance with language instructions, according to Fan.

GR00T N1.5 is set to make its debut at Computex and will be released as open source by June 9. As for Gr00t Dreams, Nvidia is still fine-tuning the timeline but aims to open-source as much as possible, Fan added.

Early adopters of GR00T N include AeiRobot, Foxlink, Lightwheel, and NEURA Robotics. AeiRobot is using the model to enable ALICE4 to understand natural language instructions and execute complex pick-and-place workflows in industrial settings. Foxlink Group is harnessing it to improve industrial robot manipulator flexibility and efficiency, while Lightwheel is validating synthetic data for faster humanoid robot deployment in factories. NEURA Robotics is exploring the model to speed up its development of household automation.

New Robot Simulation and Data Generation Frameworks Accelerate Training Pipelines

Creating highly skilled humanoid robots isn't just about the hardware; it's about feeding them a massive amount of diverse data, which can be costly and time-consuming to collect and process. Plus, testing robots in the real world comes with its own set of challenges and risks.

To bridge these gaps, Nvidia introduced several simulation technologies:

• Nvidia Cosmos Reason, a new WFM that uses chain of thoughts reasoning to help curate accurate, high-quality synthetic data for physical AI model training, now available on Hugging Face.
• Cosmos Predict 2, used in GR00T Dreams, coming soon to Hugging Face with performance enhancements for high-quality world generation and reduced hallucination.
• Nvidia Isaac GR00T-Mimic, a blueprint for generating exponentially large quantities of synthetic motion trajectories for robot manipulation, using just a few human demonstrations.
• Open-Source Physical AI Dataset, now including 24,000 high-quality humanoid robot motion trajectories used to develop GR00T N models.
• Nvidia Isaac Sim 5.0, a simulation and synthetic data generation framework, now openly available on GitHub.
• Nvidia Isaac Lab 2.2, an open-source robot learning framework, which will include new evaluation environments to help developers test GR00T N models.

Foxconn and Foxlink are already using the GR00T-Mimic blueprint to accelerate their robotics training pipelines. Meanwhile, Agility Robotics, Boston Dynamics, Fourier, Mentee Robotics, NEURA Robotics, and XPENG Robotics are simulating and training their humanoid robots with Nvidia Isaac Sim and Isaac Lab. Skild AI is developing general robot intelligence with these simulation frameworks, and General Robotics is integrating them into its robot intelligence platform.

Universal Blackwell Systems for Robot Developers

Global systems manufacturers are stepping up, building Nvidia RTX PRO 6000 workstations and servers. These systems offer a unified architecture that can handle every robot development workload, from training and synthetic data generation to robot learning and simulation.

Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro have all announced Nvidia RTX PRO 6000 Blackwell-powered servers, while Dell Technologies and Lenovo have also revealed Nvidia RTX PRO 6000 Blackwell-powered workstations.

For those times when you need even more computational power to run large-scale training or data generation workloads, developers can turn to Nvidia Blackwell systems like GB200 NVL72. Available through Nvidia DGX Cloud on leading cloud providers and Nvidia Cloud Partners, these systems promise up to 18x greater performance for data processing.