NVIDIA has launched a humanoid robot universal base model, Project GR00T, along with a new computing platform, Jetson Thor, aiming to give robots smarter brains.
Zhou Meijie, Deputy Director of the Shanghai Institute of Robotics Industry Technology, analyzed that based on the data disclosed by NVIDIA, the computational power of Project GR00T is extremely robust and disruptive. NVIDIA’s layout in the field of humanoid robots extends beyond recent developments; they have previously invested in research and development in related technologies such as robotic arms and mobile platforms.
On March 18th, local time, at the NVIDIA GTC Artificial Intelligence Conference held in San Jose, NVIDIA unveiled the humanoid robot universal base model, Project GR00T, hoping to further advance its breakthroughs in robotics and embodied intelligence.
Robots powered by Project GR00T can comprehend natural language and mimic actions by observing human behavior, such as rapidly learning coordination, flexibility, and other skills. Industry experts believe this will facilitate the development of embodied intelligence.
A universal base model designed specifically for humanoid robots.
Designed specifically for humanoid robots, the universal base model announced by NVIDIA at GTC is named GR00T, short for Generalist Robot00Technology. To support Project GR00T, NVIDIA CEO Huang Renxun introduced a new computing platform, Jetson Thor, based on NVIDIA’s Thor system-level chip and utilizing the NVIDIA Blackwell architecture. It can provide 8 trillion 8-bit floating-point operation AI performance per second to run GR00T and other multimodal generative AI models. This platform represents a significant upgrade to NVIDIA’s Isaac robot platform. To assist robots in better perceiving their environment, Huang Renxun also released the Isaac Perceptor software development kit, which features state-of-the-art multi-camera visual odometry, 3D reconstruction and occupancy mapping, as well as depth perception capabilities.
NVIDIA aims to give robots smarter brains through this initiative, enabling them to make natural mimicry actions by observing human behavior, significantly enhancing coordination, flexibility, and other aspects.
According to Latest Reports, During the keynote speech at the GTC conference, NVIDIA CEO Huang Renxun demonstrated robots completing various tasks, including robotic arms for medical and logistics handling, humanoid home robots capable of household chores like making juice, robots capable of mimicking human drumming, and robots learning and simulating human dance moves from videos, some of which can understand verbal commands from humans and respond.
NVIDIA is vigorously and forcefully advancing in the robotics market. In May 2023, at the ITF World 2023 Semiconductor Conference, Huang Renxun passionately looked forward to the future, stating that the next wave of artificial intelligence would be embodied intelligence, which refers to intelligent systems capable of understanding, reasoning, and interacting with the physical world.
In March 2024, the humanoid robot startup Figure released a demo video of the robot Figure 01, enhanced by OpenAI‘s large models, which went viral on social media platforms. In the video, Figure 01 can engage in fluent conversation with humans and analyze and execute verbal commands. In the latest round of financing in February this year, Figure received investment from technology giants such as Microsoft and NVIDIA.
“2025 may be the starting year for humanoid robots with embodied intelligence.”
Huang Dongyan, a speech technology scientist from Shenzhen-based Ubtech Robotics Corp., pointed out aspects of interest in NVIDIA’s Project GR00T humanoid robot project. On one hand, Project GR00T is currently the world’s first humanoid robot base model. On the other hand, NVIDIA has updated its Isaac robot platform on the virtual simulation platform, providing developers with a powerful end-to-end platform for developing, simulating, and deploying artificial intelligence robots, supporting synchronous training and simulation of thousands of robots.
“This will accelerate the application of humanoid robots in various industries and households, indicating that 2025 may be the starting year for humanoid robots with embodied intelligence,” said Huang Dongyan.
Zhou Meijie, Deputy Director of the Shanghai Institute of Robotics Industry Technology, believes that based on the data disclosed by NVIDIA, Project GR00T possesses extremely powerful computational power and is disruptive. NVIDIA’s deployment in the field of humanoid robots is not limited to recent developments; they have previously conducted research and investment in related technologies such as robotic arms and mobile platforms. NVIDIA’s graphics card technology provides essential computational power support for humanoid robots to have embodied intelligence, which is particularly important in the application of AI technology.
“Strong computational power support is needed for reinforcement learning and model training in humanoid robots, and NVIDIA’s graphics cards provide such computational power,” Zhou Meijie acknowledged. Currently, although the development speed of the domestic robotics industry is fast, it is overall rather impetuous. “There are few products on the market that truly possess embodied intelligence due to the lack of models and data.”