NVIDIA Unveils Open Humanoid Robot Platform to Accelerate the Future of Physical AI
Artificial Intelligence & Machine Learning News
At NVIDIA GTC Taipei, NVIDIA introduced a major new initiative aimed at accelerating humanoid robotics research worldwide. The company revealed the NVIDIA Isaac GR00T Reference Humanoid Robot, an open reference platform that combines advanced humanoid hardware, onboard AI computing, and an integrated robotics software ecosystem into a single development framework.
The announcement marks one of NVIDIA’s boldest moves yet into physical AI, as competition intensifies among companies racing to build general-purpose humanoid robots capable of operating in factories, warehouses, healthcare settings, and eventually homes.
Built on the NVIDIA Isaac GR00T development platform and powered by NVIDIA Jetson Thor onboard computing, the new reference robot combines a Unitree H2 Plus humanoid body with Sharpa tactile five-finger robotic hands designed for dexterous manipulation tasks.
NVIDIA says the platform was created to solve one of the robotics industry’s biggest challenges: fragmentation. Researchers often face disconnected systems for simulation, hardware integration, AI training, data collection, and deployment. The Isaac GR00T platform aims to unify those workflows into a single open architecture.
Jensen Huang, founder and CEO of NVIDIA, described humanoid robotics as one of the next major economic revolutions driven by artificial intelligence.
“Humanoid robots will bring physical AI to the world’s largest industries, opening a multitrillion-dollar economic opportunity,” Huang said. “The NVIDIA Isaac GR00T Reference Humanoid Robot gives researchers a single, open platform to make breakthrough discoveries toward general-purpose physical intelligence.”
The humanoid robot itself represents a highly advanced research system. Standing nearly six feet tall and weighing approximately 150 pounds, the Unitree H2 chassis includes 31 degrees of freedom across the body. When combined with Sharpa’s dual tactile robotic hands, the system reaches a total of 75 degrees of freedom, allowing sophisticated human-like manipulation and movement.
The platform also integrates multi-view sensing technologies, including stereo vision cameras mounted on the robot’s head, wrist cameras for close-range object interaction, and inertial measurement systems for motion tracking and stability control.
Powering the robot is NVIDIA’s Jetson AGX Thor T5000 compute module, featuring a Blackwell GPU capable of delivering more than 2,000 FP4 teraflops of AI performance. NVIDIA says the onboard compute enables real-time inference, sensor fusion, and autonomous control directly on the robot itself.
The Isaac GR00T software ecosystem extends beyond hardware. The platform includes tools for teleoperation data capture, simulation environments through Isaac Sim and Isaac Lab, robot policy evaluation, open foundation models for humanoid reasoning, and deployment middleware built on Isaac ROS.
Importantly, NVIDIA emphasized that researchers maintain ownership and control over their own robot data, telemetry, logs, and training datasets while using the platform.
The company also revealed that several leading robotics institutions will adopt the reference robot for next-generation humanoid research, including the Stanford Robotics Center, ETH Zurich, Ai2, and the Advanced Robotics and Controls Laboratory at the University of California San Diego.
Steve Cousins, executive director of the Stanford Robotics Center, said open robotics systems remain essential for rapid innovation.
“Robotics moves fastest when researchers can build on open platforms, share code and test ideas on real machines,” Cousins said. “The NVIDIA Isaac GR00T Reference Robot gives our students and collaborators an open humanoid reference design with dexterous hands, onboard AI compute and the NVIDIA Isaac GR00T development platform for creating, comparing and sharing robot behaviors on physical hardware.”
Researchers also view the initiative as an important step toward democratizing humanoid robotics development, which has historically been dominated by highly proprietary systems available only to large corporations.
Deepak Pathak, cofounder and CEO of Skild AI, said accessible reference platforms could dramatically accelerate experimentation across the robotics ecosystem.
“A reference design lets more researchers participate in frontier humanoid research and move from ideas to experiments faster,” Pathak said. “This helps push the whole robotics research ecosystem forward.”
NVIDIA Research will also use the system internally to continue developing Isaac GR00T open models and robotics frameworks.
The NVIDIA Isaac GR00T Reference Humanoid Robot is expected to become commercially available through Unitree in late 2026. NVIDIA also plans to release Isaac GR00T workflows for the Unitree G1 humanoid robot through GitHub and Hugging Face, further expanding access to open humanoid development tools for researchers and developers worldwide.
As AI increasingly moves from digital systems into physical environments, NVIDIA’s new humanoid robotics platform could become a foundational layer for the next generation of intelligent machines.
