NVIDIA announced a major advancement in humanoid robotics with the launch of NVIDIA Isaac GR00T N1.5, the latest iteration of its open, versatile foundation model designed for humanoid reasoning and skill execution. Alongside this, NVIDIA introduced Isaac GR00T-Dreams, an innovative blueprint for creating synthetic motion data, as well as the powerful NVIDIA Blackwell systems engineered to accelerate the development lifecycle of humanoid robots.
Prominent robotics and humanoid developers such as Agility Robotics, Boston Dynamics, Fourier, Foxlink, Galbot, Mentee Robotics, NEURA Robotics, General Robotics, Skild AI, and XPENG Robotics are leveraging NVIDIA’s Isaac platform technologies to drive progress in humanoid robot innovation and deployment.
“Physical AI and robotics will bring about the next industrial revolution,” said Jensen Huang, founder and CEO of NVIDIA. “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.”
Closing the Data Gap with Isaac GR00T-Dreams
Unveiled during Jensen Huang’s keynote at COMPUTEX, NVIDIA Isaac GR00T-Dreams is a cutting-edge blueprint designed to generate expansive synthetic motion datasets—known as neural trajectories—that enable physical AI developers to teach robots complex new behaviors and adaptability in dynamic environments.
Developers first post-train Cosmos Predict world foundation models (WFMs) tailored to their robots. By inputting a single image, GR00T-Dreams produces video simulations of the robot performing tasks in varied settings. These simulations are then distilled into action tokens—compact, actionable data packets—that instruct robots on executing new tasks.
Complementing the earlier Isaac GR00T-Mimic blueprint (introduced at NVIDIA GTC), which enhances existing datasets via NVIDIA Omniverse™ and Cosmos™, GR00T-Dreams focuses on generating entirely new synthetic data for robust model training.
Isaac GR00T N1.5: Smarter, Faster Humanoid Robots
Using synthetic data generated through the GR00T-Dreams blueprint, NVIDIA Research developed GR00T N1.5, an upgrade to GR00T N1, within just 36 hours—a process that traditionally would require nearly three months of manual data collection.
GR00T N1.5 excels at adapting to new environments and workspace layouts, with enhanced object recognition capabilities through user instructions. This update substantially improves success rates in common industrial tasks such as sorting and material handling.
Early adopters include AeiRobot, Foxlink, Lightwheel, and NEURA Robotics. AeiRobot utilizes the models to empower ALICE4 with natural language understanding for intricate pick-and-place operations in industrial environments. Foxlink Group enhances manipulator flexibility and efficiency, Lightwheel accelerates humanoid deployment through synthetic data validation in factories, and NEURA Robotics integrates these models to advance household automation.
Also Read: VERSES® Launches Genius™ Commercially
Accelerating Training Pipelines with Advanced Simulation and Data Frameworks
Creating highly skilled humanoid robots demands diverse, extensive datasets—a costly and time-intensive challenge. Physical testing carries risks and expense, underscoring the need for efficient simulation.
To address this, NVIDIA introduced several cutting-edge simulation technologies:
-
NVIDIA Cosmos Reason: A new WFM utilizing chain-of-thought reasoning for generating precise, high-quality synthetic data, now accessible via Hugging Face.
-
Cosmos Predict 2: The backbone of GR00T-Dreams, soon available on Hugging Face with enhanced world generation and reduced hallucination.
-
NVIDIA Isaac GR00T-Mimic: A blueprint for massive synthetic motion trajectory generation using limited human demonstrations.
-
Open-Source Physical AI Dataset: Now includes 24,000 high-fidelity humanoid robot motion trajectories foundational to GR00T N models.
-
NVIDIA Isaac Sim™ 5.0: An upcoming open-source simulation and synthetic data framework, soon on GitHub.
-
NVIDIA Isaac Lab 2.2: Open-source robot learning software that supports new evaluation environments for GR00T N model testing.
Foxconn and Foxlink leverage the GR00T-Mimic blueprint to accelerate robotics training. Meanwhile, Agility Robotics, Boston Dynamics, Fourier, Mentee Robotics, NEURA Robotics, and XPENG Robotics employ NVIDIA Isaac Sim and Isaac Lab for simulation and training. Skild AI focuses on general robot intelligence using these tools, while General Robotics integrates them into its intelligence platform.
Universal Blackwell Systems Powering Robotics Innovation
Global system manufacturers have embraced NVIDIA RTX PRO™ 6000 workstations and servers, providing a unified architecture capable of handling all robotics development workloads—spanning training, synthetic data creation, robot learning, and simulation.
Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro have announced NVIDIA RTX PRO-powered servers. Dell Technologies, Hewlett Packard Enterprise, and Lenovo also unveiled NVIDIA RTX PRO 6000 Blackwell-powered workstations.
For large-scale training and data generation demands, developers can access NVIDIA Blackwell systems like the GB200 NVL72, available via NVIDIA DGX™ Cloud on major cloud platforms and partners, delivering up to 18 times faster data processing performance.
