NVIDIA pushes AI for the learning of humanoid robots: Project GR00T kicks off

Is called Project GR00T (Generalist Robot 00 Technology) the new artificial intelligence platform presented by NVIDIA to GTC 2024, designed to facilitate the creation of humanoid robots able to imitate movements and natural language directly by observing humans. During the presentation, a demonstration was seen with robots capable of responding credibly to inputs given by humans, demonstrating that the project is already at a good level of maturity. With this project the computing giant aims to compete with OpenAIthe ChatGPT company, which only a few days ago made headlines with its humanoid robot Figures 01. To accompany the new platform NVIDIA also presented the computational power of the new SoC (System on Chip) Jetson Thor which uses the new ones for graphic calculations Blackwell GPUsdedicated to artificial intelligence applications, which promise performance from 7 to 30 times higher compared to the previous generation.

NVIDIA has developed a platform called Isaac, which simplifies and reduces development costs for robots for industrial or commercial purposes. Among other features, NVIDIA Isaac lets you use pre-built models to train robots, run tests and simulations, and simplify the development of autonomous robots.

NVIDIA’s strategy is therefore to attract companies interested in investing in the robotics and AI sector by providing them with a suite with which to design and develop robots from scratch. Companies like Boston Dynamics, Sanctuary it’s the same AI figures have already joined the project.

Furthermore, To automate the learning ability of robots, in addition to making the Isaac suite available, NVIDIA has planned learning models able to process a multitude of files data as input. The model allows training starting from the set of data provided: for example, it is possible to request the analysis of text, images, videos or even of behavioral manifestations linked to proxemics, mimicry and gestures. After processing, the robot can be trained and use what it has “learned” to produce features unique such as movements, speech properties (even in languages) or the ability to interact with users based on the information received in input.

gr00t-foundation-model-workflow