Fresh juice


NVIDIA doubles down on robotics and AI at GTC 2024

NVIDIA kicked off its annual GPU Technology Conference with a bang, unveiling major new products and platforms designed to accelerate the development of humanoid robots and artificial general intelligence (AGI).



In a raucous keynote at the SAP Center, NVIDIA CEO Jensen Huang laid out the company's latest innovations spanning robotics, autonomous vehicles, and industrial automation. Central to the announcements was Project GR00T, an ambitious initiative to create a generalist foundation model to imbue humanoid robots with language, vision, and movement capabilities.


"Building foundation models for general humanoid robots is one of the most exciting problems to solve in AI today," Huang told the crowd of over 15,000 attendees. "With GR00T and our Isaac robotics platform, we aim to empower the world's leading roboticists to take giant leaps towards artificial general robotics."


Project GR00T, short for Generalist Robot 00 Technology, represents NVIDIA's attempt to combine the fast moving fields of robotics and foundation models like GPT-4. The multimodal AI model is trained on a massive dataset of text, images, and motion capture data to enable human-like communication, perception, and motor skills in humanoid form factors.


NVIDIA demonstrated GR00T powering a variety of humanoid robot prototypes from partners like Agility Robotics, Boston Dynamics, and Sanctuary AI. The robots could understand natural language instructions, observe human movements to mimic them, and complete physical tasks in simulated environments.


"We're excited to partner with NVIDIA to invest in the infrastructure to enable the dream of robots being a part of daily life," said Jonathan Hurst, co-founder and chief robot officer at Agility Robotics, maker of the Digit bipedal bot.


To support the robotics push, NVIDIA unveiled new hardware and software tools tailored for humanoid development. The centerpiece is Jetson Thor, a robotic computer based on the new Thor GPU architecture delivering a massive 800 teraflops of AI compute for perception, planning, and control.


On the software side, NVIDIA expanded its Isaac robotics platform with generative AI tools like Isaac Lab for training robots in simulation at scale. Another new offering called OSMO handles the complex compute orchestration needed to develop and deploy embodied AI models.


Huang also highlighted NVIDIA's "simulation-first" approach, showing off a digital twin of an entire warehouse facility with robots, workers, sensors, and automated logistics all operating in perfect synchronization. He said this AI "gym" demonstrates how modern computer graphics and simulation can accelerate automation across manufacturing, logistics, and robotics.


In the autonomous vehicle space, the company launched DRIVE Thor, its nextgen system-on-chip for powering self-driving cars, trucks, and robotaxis. Based on the new Thor GPU architecture, DRIVE Thor delivers 2X higher performance than its predecessor and will enable manufacturers to deploy autonomous driving capabilities more rapidly.


The product blitz shows NVIDIA is serious about maintaining its lead in the red hot artificial intelligence market. While AI chips from rivals like Cerebras and potential upstarts like Apple loom on the horizon, NVIDIA is leveraging its vertical software stack to entrench itself across key AI domains.


"We are truly in the era of AI, with foundational models ushering in a new age of robotics, autonomous systems, and industrial automation," said Marc Nager, an AI leadership executive at Deloitte Consulting. "NVIDIA's announcements around embodied AI models and platforms like GR00T and Thor could spark a new wave of innovation if the potential can be realized."


Several roboticists and analysts pointed out that creating human-level artificial general intelligence remains an immense challenge. General AI that can match the broad reasoning and problem-solving abilities of humans has been a long-standing goal that has proven elusive so far.


But by fusing fields like large language models, robotics, and accelerated computing, NVIDIA is betting it can pioneer the path to artificial general intelligence. And this week's torrent of product releases shows the company putting massive technical muscle behind that moonshot goal.

Share with friends:

Write and read comments can only authorized users