© 2024 New Atlas
AI & Humanoids

NVIDIA aims to speed up humanoid development with Project GR00T

By Paul Ridden
March 20, 2024
NVIDIA aims to speed up humanoid development with Project GR00T
NVIDIA founder and CEO Jensen Huang plans to put the company at the center of humanoid robot development with the launch of Project GR00T
NVIDIA founder and CEO Jensen Huang plans to put the company at the center of humanoid robot development with the launch of Project GR00T
View 2 Images
NVIDIA founder and CEO Jensen Huang plans to put the company at the center of humanoid robot development with the launch of Project GR00T
1/2
NVIDIA founder and CEO Jensen Huang plans to put the company at the center of humanoid robot development with the launch of Project GR00T
"The GR00T model takes multi-modal instructions and past interactions as input, and produces the next action for the robot to execute," the promo video explains
2/2
"The GR00T model takes multi-modal instructions and past interactions as input, and produces the next action for the robot to execute," the promo video explains

NVIDIA has announced that it's building a new AI platform for just about all of the big players in humanoid robot development, which is based on a general-purpose foundation model that's possibly named after a talking tree-like creature from the Marvel Universe.

The Generalist Robot 00 Technology project – or GR00T for short – is described as "a general-purpose foundation model for humanoid robots, designed to further its [NVIDIA's] work driving breakthroughs in robotics and embodied AI."

As you might expect for a first introduction, low-level details are somewhat lacking. It's being designed so that bots will be able to understand natural language, for the next stage of the kind of interaction with humans that was recently demonstrated to great effect by Figure's GPT-enhanced 01 humanoid.

The technology will also allow them to learn new skills by watching humans (via videos or VR), to give them the tools they need to "navigate, adapt and interact" – and work – in the real world. Reinforcement learning at scale is facilitated through an application built on the company's Isaac Sim robotics simulation platform, while training and inference workflows can be coordinated and delivered through a cloud-based service. New tools have also been added to the Isaac suite to support industrial bot development.

"The GR00T model takes multi-modal instructions and past interactions as input, and produces the next action for the robot to execute," the promo video explains
"The GR00T model takes multi-modal instructions and past interactions as input, and produces the next action for the robot to execute," the promo video explains

NVIDIA had some new hardware to announce as well. The Jetson Thor system-on-chip includes a "high-performance CPU" and a next-gen Graphics Processing Unit that can run models like GR00T thanks to a "transformer engine delivering 800 teraflops of 8-bit floating point AI performance."

Towards the end of his keynote at the GTC developer gathering in San Jose, California, NVIDIA's always entertaining Jensen Huang was joined onstage by a line of nine humanoid robots currently in development – along with a couple of Disney bots built around the company's Jetson computers and trained in the Isaac Sim robotics simulator.

The purpose of this gathering was to show that the company is working on a new AI platform for leading lights of humanoid robot development, including Boston Dynamics, Figure AI, Agility Robotics, Unitree Robotics, Apptronik, 1X Technologies, Sanctuary AI, Fourier Intelligence and XPeng Robotics.

The pace of development in the humanoid robotics space is already proving to be incredibly fast, and with GR00T and its other platforms and hardware, NVIDIA is looking to be at the center of the action. You can tune into Huang's full keynote on the company's YouTube channel, or watch the featured robotics promo below.

NVIDIA Robotics: A Journey From AVs to Humanoids

Source: NVIDIA

Tags

AI & HumanoidsNVIDIAHumanoidRobotsArtificial Intelligence
3 comments
Paul Ridden
Paul Ridden
While Paul is loath to reveal his age, he will admit to cutting his IT teeth on a TRS-80 (although he won't say which version). An obsessive fascination with computer technology blossomed from hobby into career before hopping over to France for 10 years, where he started work for New Atlas in 2009. Now back in his native Blighty, he serves as Managing Editor in Europe.

Most Viewed

Load More
3 comments
Smokey_Bear
Great project from Nvidia, it should help out a lot of robotics startups.

The only way us humans, can ever quit our job, and have a life of leisure with minimal stress, is through humanoid robots doing the work for us. We are finally on the verge of that happening. Sure, we had Honda's Asimo walking around and climbing stairs 20 years ago, but it was all programed. AI is hugely important, these systems are finally getting the brains they've always needed.
We're just a few years away from them beginning to take jobs away from people, and once UBI is in place, we should be cheering this achievement.
95% of people do not want to work nearly everyday, doing something that they don't enjoy, for a meager paycheck to be dissolved from varies bills all to have you squeak by until the next paycheck.
This shit ends (relatively) soon.
Daishi
The idea behind this project is that LLM's are already being trained on almost all human created text but humans process much more information because they are multimodal (audio, video etc.). So they build a virtual environment (in Omniverse) and lets thousands of robots loose inside it to try to learn how to interact with the virtual world, after they succeed, they copy the code out to real robots. It's ambitious and gives them the next stage of training data needed to surpass text-based models.

@Smokey_Bear Nobody is going to call you and offer you money for the rest of your life to quit your job. It is never going to happen. Everyone wants the money but nobody wants to pay the money and that's kind of the whole problem. I could go into far more detail but the wealthiest people in the world don't call their money income (so it isn't taxed) and often have corporations, LLC's, investments, and trusts in multiple countries. If taxing wealth was easy we'd already be doing it. Nobody with the luxury of deciding how much of their money to call income is going to pay UBI to anyone. All the UBI studies overlook this to a fault.
Faint Human Outline
@Daishi

Well said with the challenges around money in its current forms. On a related thought, the value of currency also follows a supply and demand, exchanges of goods and services between people. If entire sectors of the market are operated by these robotic systems, perhaps economies will see deflation. If the perceived value in currency deteriorates, it could lose all meaning and influence.