NVIDIA has released a fully open humanoid robot platform designed to accelerate research and deployment of general-purpose robots across industries. The NVIDIA Isaac GR00T Reference Humanoid Robot, announced on May 31, 2026, combines a six-foot-tall Unitree H2 Plus chassis with Sharpa Wave tactile five-finger hands and an NVIDIA Jetson AGX Thor T5000 supercomputer brain capable of 2,070 FP4 teraflops of artificial intelligence performance. By making the entire software stack available as open source, NVIDIA is removing barriers that have historically required millions of dollars in proprietary systems, potentially reshaping how universities and research teams develop robots.

Why Is NVIDIA Betting on Humanoid Robots Right Now?

The timing reflects a convergence of three technological breakthroughs that make general-purpose robots feasible for the first time. First, foundation model artificial intelligence, the same technology powering large language models, can now be applied to robot perception and motor control through Vision-Language-Action models, or VLAs. These models enable robots to generalize across tasks they were never explicitly taught. Second, hardware has matured sufficiently; actuators, force sensors, and battery technology now allow humanoid robots to walk on uneven terrain, manipulate objects with sub-centimeter precision, and operate for hours without external power. Third, market pressure is mounting: Goldman Sachs forecasts the humanoid robot market could reach $38 billion annually by 2035, while NVIDIA's leadership suggests the addressable opportunity spans trillions of dollars across the global workforce engaged in manual and physically demanding labor.

"Humanoid robots will bring physical AI to the world's largest industries, opening a multitrillion-dollar economic opportunity. The NVIDIA Isaac GR00T Reference Humanoid Robot gives researchers a single, open platform to make breakthrough discoveries toward general-purpose physical intelligence," said Jensen Huang, Founder and Chief Executive Officer at NVIDIA.

Jensen Huang, Founder and Chief Executive Officer, NVIDIA

What Makes the Isaac GR00T Platform Different?

The hardware itself is impressive, but the real innovation lies in the software. The Isaac GR00T platform provides a complete development workflow addressing every stage of robot creation, from data collection through simulation, training, evaluation, and real-world deployment. The system includes Isaac Teleop, a teleoperation interface that allows researchers to demonstrate tasks on physical robots and record high-quality training data without requiring millions of simulation steps or thousands of hours of manual demonstration. Pre-trained GR00T foundation models, released on GitHub and Hugging Face, provide a starting point for fine-tuning on specific tasks, dramatically reducing the data and compute required to adapt robots to new applications.

The physical robot itself combines several cutting-edge components. The Unitree H2 Plus chassis provides 31 degrees of freedom across body joints, with leg actuators generating up to 360 Newton-meters of torque for locomotion and arm actuators producing up to 120 N-m for manipulation. The Sharpa Wave tactile five-finger hands add 11 degrees of freedom each, bringing the total to 75 degrees of freedom across the complete system, making dexterous manipulation possible at a level previously difficult to achieve in robotics. The onboard Jetson AGX Thor T5000 compute module delivers sufficient processing power to run large multimodal neural networks in real time without cloud connectivity, with 128 gigabytes of unified memory shared between the GPU and a 14-core ARM processor.

How Are Researchers Using the Platform?

• Data Collection: Isaac Teleop enables human-in-the-loop demonstration capture, removing one of the primary bottlenecks in robot learning pipelines by allowing efficient collection of diverse, high-quality training data at scale.
• Model Fine-Tuning: Pre-trained GR00T foundation models on GitHub and Hugging Face provide a foundation for adaptation to specific tasks using comparatively small amounts of task-specific data, rather than training from scratch.
• Simulation and Testing: The platform includes virtual environment simulation capabilities, allowing researchers to test and refine robot policies before deploying them to physical hardware.
• Real-World Deployment: Onboard compute with configurable power consumption between 40 and 130 watts allows research teams to trade computational intensity against battery life, with approximately three hours of operating time at 40 watts using a 15 Ah battery.

What Does This Mean for the Robotics Industry?

The release of an open, fully integrated humanoid robot reference design marks a shift in how robotics research is conducted. Historically, developing a capable humanoid robot required proprietary end-to-end systems costing millions of dollars, limiting research to well-funded organizations. By providing the Isaac GR00T platform as open source, NVIDIA is enabling university labs and smaller research teams to participate in advancing physical AI. This democratization could accelerate the pace of progress across the field, similar to how open-source large language models democratized natural language processing research.

The announcement also reflects a broader shift in how robots are being positioned in society. While humanoid robots have traditionally been demonstrated in controlled industrial settings and research laboratories, they are increasingly appearing in cultural and entertainment contexts. In Seoul, humanoid robots recently walked alongside human models at a fashion show, dressed in designer outfits and performing synchronized choreography, signaling a transition toward presenting robots in social and cultural environments rather than purely technical ones. This cultural integration suggests that as robots become more capable, their role in society is expanding beyond factories and warehouses into spaces where they interact directly with people.

However, the path from research platform to widespread deployment remains complex. While modern humanoid robots have become significantly better at walking, balancing, and performing choreographed movements, researchers continue to face challenges involving dexterity, autonomy, perception, and natural human-robot interaction. The Isaac GR00T platform addresses some of these challenges by providing researchers with the tools and foundation models needed to tackle these problems systematically, but the transition from laboratory demonstrations to reliable real-world operation will require sustained research effort and iterative refinement.

The broader robotics landscape is also diversifying beyond humanoid designs. Duke University researchers have developed Argus, a 20-legged robot designed around mathematical symmetry rather than biological inspiration, achieving a dynamic isotropy score of 0.91, close to the theoretical maximum of 1.0. This approach suggests that for certain applications, such as search and rescue, inspection, and navigation through unpredictable environments, non-humanoid designs may prove more effective than copying human body proportions. The diversity of approaches underway indicates that the future of robotics will likely include multiple form factors optimized for different tasks and environments, rather than a single dominant design.