Nvidia has entered the race for humanoid robotics. Through a partnership with Chinese startup Unitree, the American chipmaker is developing a robotic system with an integrated reference design focused on academic research. The idea is to help research teams access advanced hardware and, more importantly, a complete software stack that can run without proprietary platforms.
From a design perspective, the Isaac GR00T reference humanoid robot is fully modular. The Nvidia Isaac GR00T platform, combined with the Nvidia Jetson Thor embedded computer, will constitute the “brains” of the design. For the body, Nvidia opted for Unitree’s six-foot H2 Plus humanoid body with five-finger Sharpa Wave touchscreen hands (developed by Dual) that will provide the robot with real dexterous manipulation capabilities. The combination of all the features certainly justifies the robot’s place among Nvidia’s coolest reveals at the Computex 2026 event.
Although it may still be a few years before humanoid robots become part of our smart homes, Nvidia’s foray into robotics could be a step in the right direction. Prioritizing research means that the reference robot can lead to a democratization of technology. According to the company’s CEO, this move is integral as physical AI opens up a significant economic opportunity across many industries. By providing these tools, Nvidia’s robot could become a central piece of the puzzle, as it could help researchers make significant progress in developing general-purpose humanoids.
What is the capacity of Nvidia’s humanoid robot?
As a comprehensive platform for humanoid research, Nvidia’s Isaac GR00T reference humanoid robot packs many elements into a single chassis. For starters, the Unitree H2’s body measures nearly six feet tall, which, aside from looking quite dashing, is capable of 22 degrees of freedom. This is quite impressive if we also add the 22 degrees of freedom offered by two Sharpa Wave hands. The arms themselves can also lift up to 33 pounds, and when combined with full-body control capabilities, including reaching and lifting, the Isaac GR00T is nothing short of capable.
In addition to its muscles, the Nvidia robot has many brains. The Nvidia Jetson AGX Thor sports a 2,070 teraflop Nvidia Blackwell GPU for maximum AI performance, with 128 GB of available memory. In addition to a front-facing camera, the robot supports a wide range of connectivity options for multi-view sensory processing. This is a powerful package, so it’s really no surprise that several institutions like the Stanford Robotics Center and ETH Zurich have decided to use the Nvidia robot for research.
Many people are afraid of new developments in robotics. Sure, a humanoid robot learning to read facial expressions on YouTube is nightmare fuel, but watching these machines learn to dance is also quite captivating, especially for tech enthusiasts. While Nvidia’s entry into robotics will likely inspire plenty of references to Skynet and Asimov, it will also help research teams build a real foundation for advancing the field. Whether this will be positive remains to be seen, but a company like Nvidia championing this technology could be a sign that humanoid robots are about to become mainstream.
