The more technology evolves, the closer it is to life.
In the past two years, AI has occupied the absolute C position on the list of hot topics in technology. Following the explosion of ChatGPT, as another important branch of AI development, "embodied intelligence" is becoming a hot topic of widespread concern around the world. Suddenly, "embodied robots" have also become a new vane in the science and technology industry.
According to the definition of CCF experts, embodied intelligence refers to an intelligent system based on the perception and action of the physical body, which obtains information, understands problems, makes decisions and realizes actions through the interaction between the agent and the environment, so as to produce intelligent behavior and adaptability, and its key word is "embodiment", that is, it is not just abstract algorithms and data, but interacts with the world through physical forms.
The so-called embodied robot, that is, the first-person intelligence with initiative, is essentially a robot that can interact with the environment and perceive, and can independently plan, make decisions, act, and have the ability to execute. Not just humanoid robots, any tangible intelligent machine that can move through space is a form of embodied robot, the core goal of which is to be able to hear human language, then break down tasks, plan subtasks, interact with the environment, and finally complete the corresponding tasks.
Judging from the current development status of the industry, humanoid robots are the best bearing form of embodied intelligence. Driven by policy, capital, and the market, global technology giants such as NVIDIA, OpenAI, Microsoft, and Amazon have re-injected into the market, and the technology circle seems to have released a signal of the next outlet-the era of humanoid robots is approaching.
Increase the robot track and lead the new wave of humanoid robots
As a global leader in AI computing power, NVIDIA's layout in the field of intelligent robots mainly revolves around the three core areas of large models, data, and development platforms, with the ultimate goal of building an underlying development ecosystem for robots.
Since 2018, NVIDIA has been working on intelligent robots. At that time, NVIDIA launched NVIDIA Isaac, a new hardware, software, and a virtual world robot simulator, Jetson Xavier, a computer platform designed specifically for robotics, and related robotics software toolkits, firing the first shot into the field of intelligent robotics.
From 2019 to 2022, NVIDIA opened up an all-round layout in the field of intelligent robots. These include the Isaac SDK SDK in 2019, the Omniverse platform in 2021, and Isaac Nova Orin in 2022.
In 2023, NVIDIA has released a new round of "generative AI + robotics" blueprints, making a strong foray into the field of intelligent robotics. Leveraging the latest generative AI and advanced simulation technologies, NVIDIA has announced the new Jetson AGX Orin industrial-grade module and new versions of Isaac ROS and Isaac Sim software, enhancing performance across the product line while lowering the barrier to access.
When it comes to embodied intelligence, NVIDIA founder Jensen Huang's "platform" has never stopped.
Huang has repeatedly said in public speeches that "the next wave of AI will be embodied intelligence." At this year's GTC24, Huang also said that "developing a basic model for a general-purpose humanoid robot is one of the most exciting topics in the AI space today."
In February, NVIDIA also established GEAR, a general-purpose embodied agent research lab, which focuses on research in the fields of multimodal models, general-purpose robots, virtual world-based agents, simulation, and synthetic data to promote intelligent applications across multiple modalities and scenarios.
It is worth mentioning that at GTC24, NVIDIA launched the GR00T (Generalist Robot 00 Technology) and Jetson Thor, a new humanoid robot computer, which not only marks a major breakthrough in the field of humanoid robot technology for NVIDIA, but also sets a new benchmark for the entire robotics industry.
In detail, the core functions of GR00T include natural language understanding, behavior imitation and multimodal learning, which can effectively change the way humanoid robots learn the real world: by understanding natural language text, human language, videos, live demonstrations, etc., the robot can imitate human body movements or other behaviors, so as to achieve interaction with the real world.
Jetson Thor is NVIDIA's new computing platform designed specifically for humanoid robots, which is the "brain" of humanoid robots.
The computing platform behind it is critical to achieve efficient humanoid robots, and Jetson Thor is the driving force behind the GR00T.
Capable of performing complex tasks and interacting safely and naturally with humans and machines, Jetson Thor has a modular architecture optimized for performance, power consumption, and size, and can be easily integrated into various types of humanoid robots, supporting the overall performance of the robot and greatly simplifying the design and development of humanoid robots.
What's more, Jetson Thor brings AI capabilities capable of driving humanoid robots. Continuing the legacy of the Jetson series' high performance, Jetson Thor delivers 800 trillion 8-bit floating-point AI performance per second to run multimodal generative AI models such as the GR00T, powered by the next-generation NVIDIA Blackwell architecture and Transformer engine.
Compared to NVIDIA's previous-generation Hopper architecture, the Blackwell architecture delivers several times the performance improvement in every aspect. In addition, new precision has been added to the Blackwell Tensor cores, and the second-generation Transformer engine built on this architecture supports 4-bit floating-point AI inference, doubling the compute power and model size.
In addition, Jetson Thor integrates a functional safety processor, a high-performance CPU cluster, and 100GB of Ethernet bandwidth to deliver superior performance, whether it's real-time perception and decision-making in complex environments, or stable operation under high loads.
The three-layer computing platform hits the market and accelerates the development of the global humanoid robot industry
"Processor + Platform + Development Tools" is the "three-in-one" that NVIDIA has to enter every new market, and the same is true for humanoid robots.
Different from traditional automated machines or simple remote-controlled devices, humanoid robots have more complex body structures and perception capabilities, and can interact with the environment more naturally and flexibly. The emergence of this type of robot marks the leap of AI from a single data processing to a comprehensive perception, decision-making, and action capability. Behind all of this, high-performance computing power is the backbone of NVIDIA's strength.
As NVIDIA's flagship product in the field of AI, the NVIDIA AI supercomputer is a super-intelligent platform that integrates high-performance computing, deep learning, and big data analysis.
In the development process of humanoid robots, model training is a crucial part, which needs to deal with massive amounts of data and complex algorithms. The NVIDIA AI supercomputer, with its powerful GPU-parallel computing capabilities, can significantly accelerate this process, shorten the model training cycle, and improve training efficiency, which makes it possible to rapidly develop and iterate humanoid robots. In addition, through efficient model training, humanoid robots can also understand the environment more accurately and execute instructions, thus demonstrating more powerful capabilities in practical applications.
In addition to its computing power, NVIDIA has many years of experience in simulation, and its complete solution for robotics, NVIDIA Isaac, has received a lot of attention since its inception.
The Isaac platform features a suite of NVIDIA acceleration libraries, AI foundational models, and advanced simulation technologies that robot manufacturers can integrate into their technology stacks.
Isaac Sim:Build a seamless bridge between the virtual and the real
Isaac Sim is an NVIDIA-built robotics simulation reference application and one of the core components of NVIDIA Isaac, with key features such as a highly realistic robot simulation environment, modular design and flexible application, support for customization and extension, and seamless integration with other NVIDIA tools.
Built on the Omniverse platform, Isaac Sim integrates Omniverse's advanced technologies and tools, such as the PhysX physics engine, real-time ray tracing, path tracing, and Material Definition Language (MDL), to provide a highly realistic, interactive virtual environment for the design and testing of humanoid robots.
In this virtual world, developers can simulate a variety of complex scenarios, from simple object grabbing to complex environment exploration, without actually building a physical prototype, to fully evaluate the behavior and performance of the robot. This not only greatly reduces R&D costs, but also accelerates the iteration cycle, enabling humanoid robots to adapt to diverse application scenarios more quickly.
The power of Isaac Sim is also that it integrates an advanced physics engine and AI algorithms, which can simulate near-real physical interactions, such as collisions, friction, gravity, etc., providing a training ground close to the real world for robots.
In addition, Isaac Sim can be seamlessly integrated with ROS2, making it easy for developers to connect their robot's brain to the virtual world for efficient development and testing, all of which are driving the development and adoption of robotics simulation technology.
Isaac Lab：The intelligent cradle of humanoid robot learning
If Isaac Sim is the "training ground" for humanoid robots, then Isaac Lab is the "intelligent cradle" of humanoid robots. Isaac Lab is a lightweight application built on top of Isaac Sim that is optimized for reinforcement, mimicry, and transfer learning to train all types of robot instances. Here, developers can use advanced technologies such as reinforcement learning and imitation learning to teach robots how to perceive their environment, make decisions, and perform tasks.
Isaac Lab is tightly integrated with Isaac Sim, forming a complete closed loop from simulation to learning. Currently, Isaac Lab can support simultaneous training and simulation of thousands of robots.
The launch of NVIDIA's Layer 3 computing platform marks a new phase of development for the humanoid robotics industry. Together, the NVIDIA AI supercomputer, the NVIDIA Isaac Sim simulation platform, and the NVIDIA Jetson Thor humanoid robot computer form a complete ecosystem for the development, testing, and application of humanoid robots. The establishment of this ecosystem not only accelerates the development and optimization of humanoid robots, but also provides infinite possibilities for the application of humanoid robots in intelligent manufacturing, commercial services, medical health, home services and other fields.
With the continuous advancement of technology and the continuous expansion of the market, the humanoid robot industry is expected to usher in a more brilliant tomorrow, and the NVIDIA Layer 3 computing platform will be an indispensable innovation engine and driving force in this process.
Epilogue and the future
In today's rapid development of science and technology, humanoid robots are gradually becoming a bridge connecting reality and the future, they not only carry the yearning of human beings for intelligent life, but also an important symbol of the scientific and technological progress of the whole society.
With its deep technology accumulation and forward-looking strategic vision, NVIDIA is gradually building a unique ecosystem in the field of humanoid robots. From providing a high-performance computing platform, to creating a simulation environment, to supporting the actual operation of models, NVIDIA's all-round, multi-layered layout not only demonstrates its unwavering confidence in the future of humanoid robots, but also injects strong impetus into the development of the entire industry.
In addition, at GTC24, the finale is also 9 humanoid robots. It can be seen that today, robots have been placed in an important position in NVIDIA's business map, and the era of embodied robot transformation belonging to NVIDIA has begun.