NVIDIA Unveils the Ultimate Brain for Embodied Intelligence
NVIDIA Jetson AGX Thor unlocks the ChatGPT moment for robotics, powering general-purpose robots via simulation-first training, world models & edge AI.
“RoboPub” Publication: 20% Discount Offer Link.
“We are moving from an era of perceptual intelligence to a new era of action intelligence.”
This is the next pivotal node in the robotics era as foreseen by Professor Fei-Fei Li, co-director of Stanford HAI and a pioneer in embodied intelligence. She believes that the next challenge for robots is not about seeing more accurately, but about making correct decisions and taking appropriate actions based on what they see — a capability that requires a completely new, generalised AI framework.
Over the past several decades, robots have been rigidly confined to fixed workspaces, performing precise yet highly repetitive single tasks. Today, driven by breakthrough advancements in AI technology — represented by large models — the global robotics industry is ushering in a historic “singularity moment” under new concepts such as embodied intelligence. This marks a paradigm shift from specialised to general-purpose systems.
People are beginning to expect that robots will no longer be tools custom-built for specific assembly lines, but universal intelligent agents capable of adapting to complex, unstructured environments and performing a wide variety of tasks, or what we call General-Purpose Robots.
How can robots accelerate toward the critical point of “generalisation”?
To realise this grand vision of generalisation, the industry has placed unprecedented and stringent demands on underlying supporting technologies. Four major technical pillars appear indispensable.
Training a universal robotic “brain” that can understand the ever-changing physical world requires processing visual, linguistic, and motion data on a scale far exceeding anything seen before. This demands a leap in computing power from today’s thousand-card clusters to ten-thousand-card or even larger-scale systems.
As NVIDIA founder and CEO Jensen Huang has repeatedly emphasised: “The ChatGPT moment for the robotics era is coming, and it must be built on the foundation of accelerated computing. This kind of brain model requires unprecedented computational density.”
At the same time, training robots in the real world is prohibitively expensive and inefficient. High-fidelity simulation platforms have thus become an essential “training ground.”
