Inside the NVIDIA Isaac GR00T Reference Humanoid Robot: The Open Platform Designed to Accelerate Humanoid Development

• The Body: Unitree provides the H2 Plus humanoid chassis, a bipedal robot standing nearly 6 feet tall and weighing approximately 150 lbs. It features 31 degrees of freedom in the body, with arm torque of up to 120 Nm, leg torque up to 360 Nm, and a payload capacity of 7 kg rated and 15 kg at peak . The system is completed by Sharpa’s Wave five-fingered tactile hands, which add 22 degrees of freedom per hand, bringing the total to 75 across the entire robot .

• The Brain: The onboard intelligence is powered by the Nvidia Jetson Thor module, which includes a Blackwell GPU delivering 2,070 FP4 TFLOPS of AI performance, a 14-core Arm CPU, and 128GB of memory . This runs the NVIDIA Isaac GR00T open development platform, a comprehensive software stack that spans data capture and generation, synthetic motion data creation, model training, evaluation, and final real-world deployment .

Who Is Behind the H2+ Design

The project brings together three companies with distinct but complementary expertise.

• Nvidia contributed the Jetson Thor compute module, the Isaac GR00T software platform, and acted as the overall system integrator .
• Unitree Robotics, based in Hangzhou, China, provided the H2 Plus humanoid robot body . The announcement notably coincided with the same day Unitree went public .
• Sharpa, a robotics company from Singapore, supplied the Wave five-fingered hands, which are critical for allowing the robot to perform dexterous manipulation tasks .

How the Platform Aims to Accelerate Innovation

The core value proposition of the Isaac GR00T reference design is the unification of a traditionally fragmented workflow. Robotics labs typically must source hardware, integrate it with custom software, set up data collection pipelines, build simulation environments, train models, and then deploy them—a process that can take months before any meaningful research begins .

The reference design collapses this timeline by delivering a known-working system. Because the body and brain are pre-integrated, researchers can bypass hardware bring-up and move directly to developing and validating new skills .

This open standard approach also democratizes access to frontier humanoid hardware. Research institutions that previously lacked the resources to build or purchase proprietary platforms can now use the same advanced tools . The Isaac GR00T platform itself provides a complete workflow that includes the ability to generate synthetic motion data—virtual training scenarios that help robots learn new behaviors and adapt to changing real-world conditions .

Several leading research institutions have already committed to using the platform for their work, including the Allen Institute for AI (Ai2), ETH Zurich, the Stanford Robotics Center, and the Advanced Robotics Lab at UC San Diego . Shipments for these early partners are expected to begin in October 2026 .