Humanoid robots are becoming ever more practical. At GTC developer conference, Nvidia entered the fray with Project GR00T (pronounced Ground Robot).
Training its humanoid embodiment using GPU-accelerated simulations from NVIDIA, it uses imitation learning from just a handful of demonstrations to acquire skills through imitation. Isaac Lab provides reinforcement learning while OSMO serves as compute orchestration service to generate robot movements from video data.
What is Project GR00T?
Nvidia CEO Jensen Huang introduced Generalist Robot 00 Technology (GR00T), their latest effort in robotics and “embodied AI,” during his keynote at 2024 GPU Technology Conference (GTC). Nvidia says GR00T allows robots to understand natural language while mimicking movements by watching others perform specific actions – this should allow them to gain coordination, dexterity and other necessary skills needed to adapt and interact with real-world situations according to Nvidia.
Nvidia demonstrated multiple humanoid robots powered by its GR00T chip during its presentation, from companies including Agility Robotics, Apptronik, Fourier Intelligence and Unitree Robotics – performing various tasks and intended to demonstrate both its power as a learning device as well as safer human-robot interaction.
Nvidia’s Jetson Thor computing platform provided power for these demonstrations, designed specifically to address the complex tasks required of humanoid robots. Based on Nvidia’s Blackwell architecture, this system-on-chip can deliver up to 800 Teraflops of AI performance for humanoid robots; providing them with enough processing power to run multimodal generative models and other AI systems.
Nvidia unveiled not only their latest computer, but also a suite of software and hardware for humanoid robots: Isaac Manipulator and Isaac Perceptor as well as pretrained models, libraries, and reference hardware that will allow developers to create more powerful, faster, safer robots.
Nvidia recently unveiled an upgrade to their Isaac robotics platform that will enable humanoid robots built on GR00T to use generative AI for learning and adaption. This system includes Isaac Lab for reinforcement learning, OSMO compute orchestration service and tools for creating foundation models for humanoid robots; in addition, Isaac Arms improve robot arm performance while Isaac Perceptor handles multi-camera 3D computer vision.
Discover the best artificial intelligence courses, click here!
What is GR00T’s goal?
Nvidia is hoping to bolster the abilities of humanoid robots through Project GR00T, its foundation model for Generalist Robot 00 Technology (not related to Marvel’s evil talking space tree). This innovative model helps robots understand natural language, mimic movements by watching actions and quickly learn coordination, dexterity and other skills essential for navigation, adaptation and interaction in real world situations. CEO Jensen Huang demonstrated these benefits during a keynote at GTC developer conference and demonstrated some GR00T robots completing tasks that would normally challenge other robots. During his keynote at GTC CEO Jensen Huang displayed these robots performing tasks that other robotic technologies would struggle with.
Nvidia Isaac robotics platform provides developers with an all-encompassing toolkit for designing, simulating and deploying AI-powered robots. This includes GPU-accelerated reinforcement learning in Isaac Lab; OSMO compute orchestration service manages real and synthetic robot development workflows; as well as pre-trained models called Isaac Manipulator and Isaac Perceptor that offer faster robot arm performance and enhanced multi-camera 3D computer vision respectively.
Nvidia is providing its own computational horsepower to make GR00T come to life, using Jetson Thor as the humanoid computer of choice. Equipped with its Blackwell graphics architecture and transformer engine capable of providing 800 Teraflops of 8 Bit AI Performance, Jetson Thor allows simulation workflows and generative AI models such as GR00T to run seamlessly, according to Nvidia in an in-person briefing session.
Nvidia is positioning their platform as the perfect computing environment for humanoid robots, partnering with Figure AI, 1X Technologies and Agility Robotics to offer solutions based on it. In addition to supporting GR00T from Apptronik’s Apollo robot using Nvidia technology, Apollo will learn new tasks by rapidly observing human demonstrations rather than complex programming alone – all part of Nvidia’s “embodied AI” strategy of creating powerful computers with artificial intelligence that can perceive, comprehend and interact with reality in real-world settings.
Discover the best artificial intelligence courses, click here!
What is GR00T’s technology?
Project GR00T, or Generalist Robot 00 Technology as it is commonly known, will enable robots to learn from humans by imitating their movements by watching humans move. This way, robots will quickly gain coordination, dexterity, adaptability and other essential skills necessary for navigation in real world situations – without years of rigorous training sessions or risking an uprising from fellow machines.
Nvidia CEO Jensen Huang showcased how GR00T could help humanoid robots be smarter and more functional during his GTC 2024 keynote, showing an example wherein one powered by this technology learned how to move its head and arms by watching humans perform those movements before replicating those actions autonomously. According to Nvidia, such learning will enable more functional yet cost-efficient robotics systems.
Nvidia recently unveiled Jetson Thor, a system-on-chip optimized to meet performance, power, and size requirements of humanoid robot development powered by GR00T-based humanoids. This computer features NVIDIA’s Blackwell architecture GPU as well as an AI transformer engine delivering 800 Teraflops of performance for easier robotic design processes for robotic developers.
NVIDIA Isaac, a robotics platform designed to provide an end-to-end development and simulation ecosystem for robot embodiment, includes NVIDIA Isaac Lab for reinforcement learning as well as OSMO compute orchestration service which offers training simulation and hardware-in-the-loop validation services. Furthermore, Isaac platform has also been enhanced with new modules like Isaac Manipulator which offers fast robot arm performance as well as Isaac Perceptor which offers multi camera 3D vision capabilities.
Nvidia claims its GR00T and Isaac platforms will usher in an age of embodied AI where smart, adaptive robots become ubiquitous. Nvidia hopes developers will use its software, tools and hardware to create various kinds of robots suitable for various uses – using open source algorithms as well as APIs that make developing software for specific robots or tasks simpler.
Discover the best artificial intelligence courses, click here!
What is GR00T’s impact?
Nvidia’s Project GR00T stands out in the field of robotics and embodied AI (AI that can physically integrate itself into physical environments), as a multimodal generative foundation model will serve as the brain for robots able to understand human language and mimic movements by watching people – quickly learning coordination, dexterity and other skills necessary for adapting and interacting with real life environments. At his GTC keynote presentation, CEO Jensen Huang demonstrated humanoid robots powered by GR00T performing various tasks successfully during his keynote presentation of humanoid robots powered by GR00T performing various tasks successfully during his GTC keynote address.
Nvidia has designed Jetson Thor, a system-on-chip designed specifically to help build humanoid robots leveraging GR00T, as a computer specifically for them to use during development of such humanoid robots. Engineered to manage complexity while providing safe and intuitive human-robot interaction, its modular architecture optimizes for performance, power and size while boasting 800 Teraflops of 8 Bit AI performance – enough power for running multimodal generative models such as GR00T.
Nvidia also announced updates to their Isaac platform that will make designing and deploying AI-powered robots easier, including new modules such as Manipulator and Perceptor that provide state-of-the-art dexterity and modular AI for robotic arms and mobile robots, and Isaac Lab which enables fast parallel simulation in GPU accelerated virtual environments.
Figure AI, Agility Robotics, Boston Dynamics, Apptronik and Sanctuary AI have already joined forces with Nvidia to integrate its technology into their products. Nvidia will showcase two such partnerships at this show; these advancements should accelerate deployment of AI into real life work settings more rapidly and provide access to it for more people.
Are you ready to have your fleet of humanoids robots? Contact us today!
