End-to-end humanoid development with Isaac GR00T & Unitree G1 | Robotics Office Hour
Join us for a walkthrough of the NVIDIA Isaac GR00T End-to-End workflow for humanoid robots. In this session, we break down how to develop and deploy a manipulation policy on the Unitree G1, from environment setup all the way to robot deployment. We'll walk through the full reference workflow at a conceptual level, showing how the pieces fit together. Whether you're working fully in simulation or on the physical robot, we'll cover how the workflow supports these two independent paths, and how you can adapt the same steps to your own robot and task. Follow along with the full workflow docs: https://docs.nvidia.com/learning/physical-ai/gr00t-e2e-workflow/latest/index.html ---------------------------------------------------------------------------- 0:00 Welcome and NVIDIA Isaac GR00T Workflow 9:03 End-to-End Humanoid Development Platform 12:17 Isaac Lab-Arena and Real Robot Setup 14:26 Isaac Teleop Demonstration Data Collection 15:21 GR00T N1.7 Vision-Language-Action Training 18:03 Isaac Lab-Arena Policy Evaluation 20:40 Isaac ROS Deployment on Jetson Thor 24:42 Teleop Dataset Size and Data Augmentation 27:18 Q&A: Real-World Robot Safety 33:43 Unitree G1 Success Rates 40:43 Agile Whole-Body Controller and Sonic 53:45 Industrial Humanoid Deployment Gaps ---------------------------------------------------------------------------- List of Resources - End-to-End Physical AI With the Unitree G1 guide - main workflow guide (https://docs.nvidia.com/learning/physical-ai/gr00t-e2e-workflow/latest/index.html) - Real robot workflow hardware specs - hardware setup reference (https://docs.nvidia.com/learning/physical-ai/gr00t-e2e-workflow/latest/real-robot-workflow/real-overview.html) - GR00T fine-tuning simulation context - training context and GPU requirements (https://docs.nvidia.com/learning/physical-ai/gr00t-e2e-workflow/latest/simulation-workflow/groot-fine-tuning-sim.html#confirm-training-context) - NVIDIA Brev - cloud compute option mentioned for training (https://brev.nvidia.com/) - CloudWalk Research GR00T N1.7 + Unitree G1 collection - community example shared in chat ( https://huggingface.co/collections/cloudwalk-research/gr00t-n17-unitree-g1-grab-the-bottle-sonic-wbc) - NVIDIA Physical AI Learning catalog - related Isaac, GR00T, and Jetson learning paths (https://docs.nvidia.com/learning/physical-ai/) - Isaac ROS setup on Jetson Thor - deployment environment setup (https://docs.nvidia.com/learning/physical-ai/gr00t-e2e-workflow/latest/real-robot-workflow/isaac-ros-setup.html) - NVIDIA open humanoid robot reference design - broader platform context (https://nvidianews.nvidia.com/news/nvidia-open-humanoid-robot-reference-design) Continue the discussion and ask follow-up questions in the NVIDIA Omniverse Discord thread: https://discord.com/channels/827959428476174346/1525285331446796318 ---------------------------------------------------------------------------- 📆 Check out the full calendar for all of our upcoming events → https://nvda.ws/3JqaWnA ---------------------------------------------------------------------------- ⬇️Get Started → https://nvda.ws/4cZAZO1 👀Explore OpenUSD → https://nvda.ws/3CeozBQ 👥Join the Community → https://nvda.ws/3ZMfc6e #NVIDIA #NVIDIAOmniverse #IsaacGR00T #IsaacSim #IsaacLab #IsaacROS #JetsonThor #UnitreeG1 #PhysicalAI #HumanoidRobots #Robotics #RobotLearning #Simulation