Nvidia announced a “moonshot” to create human-level AI embodied in robot form

In sci-fi films, the rise of humanoid artificial intelligence often goes hand in hand with physical platforms, such as androids or robots. While the most advanced AI language models to date may sound like disjointed voices echoing from an anonymous data center, they may not stay that way for long. Some companies like Google, Figure, Microsoft, Tesla, Boston Dynamics, and others are working to give AI models a body. It’s called “sculpture,” and AI chipmaker Nvidia wants to speed up the process.

“Building foundation models for general humanoid robots is one of the most exciting problems to solve in AI today,” Nvidia CEO Jensen Huang said in a statement. Huang spent part of a keynote at Nvidia’s annual GTC conference on Monday promoting Nvidia’s robotics efforts. The next generation of robotics will likely be humanoid robotics, Huang said. “We now have the technology necessary to envision normal humanoid robotics.”

To that end, Nvidia announced Project GR00T, a general-purpose foundation model for humanoid robots. As a kind of AI model itself, Nvidia hopes that GR00T (which stands for “Generalist Robot 00 Technology” but sounds like a very popular Marvel character) will serve as an AI mind for robots, which It will enable them to learn skills and solve different tasks. Bee. In a tweet, Nvidia researcher Linksey “Jim” Fan called the project “our moonshot for solving embodied AGI in the physical world.”

AGI, or artificial general intelligence, is a poorly defined term that usually refers to a hypothetical human-level AI (or beyond) that can learn any task that a human can do without special training. Given a capable enough human body powered by AGI, one can imagine fully autonomous robotic assistants or workers. Of course, some experts believe that true AGI is a long way off, so it’s possible that Nvidia’s goal is more aspirational than realistic. But that’s also what makes Nvidia’s plan a moonshot.

“The GR00T model will enable a robot to understand multimodal instructions, such as language, video, and demonstrations, and perform a variety of useful tasks,” Fan wrote on X. “We are collaborating with many leading humanoid companies around the world, so that GR00T can be transferred to sculptures and help ecosystems flourish.” We reached out to Nvidia researchers, including Fan, for comment but did not receive a response by press time.

Nvidia is designing the GR00T to understand natural language and mimic human movements, potentially allowing the robot to learn coordination, dexterity and other skills to navigate and interact with the real world like a person. It is necessary to communicate with them. And as it turns out, Nvidia says that humanizing robots could be the key to creating functional robot assistants.

Humanoid key

So far, we’ve seen many robotics platforms that aren’t humanoid, including robot vacuum cleaners, automated lawn mowers, industrial units used in automobile manufacturing, and even research arms that fold laundry. can. So why focus on imitating the human form? “In a way, humanoid robotics is potentially simpler,” Huang said in his GTC keynote. “And that’s because we have a lot of simulated training data that we can provide the robot, because that’s exactly how we’re built.”

This means researchers can feed training data samples from human movements into AI models that control the robots’ movements, teaching them how to move and balance better. be established. Also, humanoid robots are particularly convenient because they can fit anywhere a person can, and we designed a world of physical objects and interfaces (such as tools, furniture, stairs, and appliances). which can be used according to the human form.

Alongside the GR00T, Nvidia also debuted a new computer platform called Jetson Thor, based on Nvidia’s Thor system-on-a-chip (SoC), as part of the new Blackwell GPU architecture. , which he hopes will underpin this new generation of humanoid robots. . The SoC reportedly includes a Transformer Engine capable of 800 teraflops of 8-bit floating-point AI computation to drive models like the GR00T.