NVIDIA demonstrates project GR00T at the GTC conference: aiming to provide multi-modal AI technology support for future humanoid robots

BlockBeats news, on March 19, according to VentureBeat, Nvidia demonstrated its latest project GR00T at its annual GTC developer conference, a multi-modal artificial intelligence platform designed to power future humanoid robots. Project GR00T utilizes a common base model that enables humanoid robots to process a variety of inputs such as text, speech, video, or even live demonstrations to perform specific general actions. The project was developed with tools from Nvidia’s Isaac robotics platform, including the newly launched Isaac Labs for reinforcement learning. To ensure that humanoid robots can run complex multi-modal models like the GR00T, Nvidia also launched the Jetson Thor computing platform designed for humanoid robots. Based on the company's Thor SoC, the computer includes a high-performance CPU cluster and a next-generation GPU based on Nvidia's Blackwell architecture with a Transformer engine that delivers 800 teraflops of 8-bit floating-point AI performance. With its Isaac robotics platform, Nvidia provides a powerful end-to-end platform for the development, simulation and deployment of AI-driven robots. Project GR00T is brought to life through the newly launched Isaac Labs based on Isaac Sim, and the OSMO compute orchestration service for managing training and simulation workloads in parallel on Nvidia DGX and Nvidia OVX. This set of Nvidia advancements is expected to be available in the second quarter of this year, with Project GR00T currently still in early access. Nvidia is accepting applications to open the technology to more humanoid robot developers, but a timeline for a wider public release is currently unclear.