Within the framework of Asia's largest technology exhibition COMPUTEX 2025 taking place in Taiwan (China), NVIDIA CEO Jenseng Huang gave a presentation that attracted thousands of experts and technology enthusiasts like every year. Besides introducing chip products, the head of NVIDIA spent a lot of time announcing the latest applications of AI in Robot development.
Regarding Agents AI, the CEO believes that robots are the most widely used products when they need to know how to "perceive, understand, plan" and "learn how to become robots". To do this, robots need to be tested and trained, but the real environment is not enough due to limited resources and time.
The Newton architecture with NVIDIA Warp is the underlying technology, running in various simulators such as MuJoCo Playground and Isaac Lab. (Image: NVIDIA)
Robot simulations need to be built on fundamental physics rules to predict how robots will behave in different situations and environments. However, simulations often do not match reality. Robot developers need a unified solution to test and validate robot control algorithms, design prototypes quickly and cost-effectively.
Since then, a "virtual world " has been developed by NVIDIA with Disney Research and Google DeepMind. This platform is called Newton, built based on the rules of physics with high fidelity, simulated in real time. Newton is also open source for the community, according to the plan to launch in July.
Also when training robots, companies often have to have human workers perform the movements that the robots "imitate". The large workload and time constraints are also the reason why NVIDIA introduced the Groot Dreams platform. From "dreams" - 2D images inferred by AI on demand - robots can learn the best synthetic version and perform in a 3D environment.
AI image predicts according to command (left) and robot executes (right). (Photo: NVIDIA)
Also during his presentation, the NVIDIA CEO revealed plans to sell an interconnect technology to speed up communication between chips, which is essential in building and deploying artificial intelligence tools.
The company has launched a new version of its NVLink technology, called NVLink Fusion, and will sell it to other chip designers to help build powerful custom AI systems with multiple chips linked together.
Nvidia said Marvell and MediaTek plan to adopt NVLink Fusion technology for their custom chip development efforts. Other partners include Alchip, Fujitsu and Qualcomm.
Nvidia’s chips, however, face an uncertain outlook in China. In an interview with Stratechery, Huang said the company “missed out on $15 billion in revenue” in China after the U.S. imposed restrictions on shipments of its H20 chips. The company said last month it would take a $5.5 billion charge related to the restrictions.
COMPUTEX 2025 will take place from May 20-23 at the Taipei Nangang Exhibition Center. With the theme "AI Next", this year's exhibition will attract 1,400 exhibitors, bringing together the world's leading technology corporations to bring cutting-edge breakthroughs in the field of artificial intelligence.
Source: https://vtcnews.vn/nvidia-bat-tay-disney-google-tao-the-gioi-ao-cho-robot-hoc-cach-nhan-thuc-ar944000.html
Comment (0)