Nvidia and Unitree share a blueprint for a humanoid robot

In short: Nvidia says researchers can build and train a humanoid robot using Unitree’s H2 Plus body, Nvidia’s Thor chip, a dexterous hand, and new software.

What happened

Nvidia CEO Jensen Huang announced a new “blueprint” for a humanoid robot. Think of it like a recommended parts list and setup guide, so labs can start faster instead of building everything from scratch.

The blueprint centers on Unitree’s H2 Plus, a roughly 6-foot-tall, 150-pound humanoid robot made by a Chinese robotics company. Nvidia pairs it with its Thor T5000 chip, which is the robot’s “brain” (the computer that runs the AI software). Nvidia also includes software meant to make it easier to program the robot and train it to do tasks.

The design also highlights a more humanlike robot hand from Singapore-based Sharpa. The hand can do delicate actions like peeling an apple, which matters because handling small objects is still a hard problem for robots.

Nvidia’s Spencer Huang, who leads product work for robotics, told WIRED the goal is to provide Nvidia chips and software to many humanoid robot makers. He said Unitree is first, but not the last, and described the approach as combining “the best of both worlds.”

Why it matters

This partnership sits in the middle of growing US and China competition in technology. Some US politicians have proposed banning Chinese humanoid robots in government use, and past research raised concerns that some Unitree robots could capture and transmit data. Nvidia says this blueprint includes security features aimed at protecting users’ data and AI models, which could be important for universities and labs deciding what robots to buy and study.

Source: Wired