NVIDIA Isaac™ GR00T is a research initiative and development platform for creating general-purpose robot foundation models and data pipelines. Built on NVIDIA Omniverse™ and Cosmos™, it aims to accelerate humanoid robotics research and development by providing the necessary tools and infrastructure. Isaac GR00T features robot foundation models for cognition and control, simulation frameworks built on NVIDIA Omniverse™ and Cosmos™, data pipelines for generating synthetic data and environments, and a computer in the robot—Jetson AGX Thor™—to run the entire robot stack.

Features

• Robot Foundation Models: Includes models for cognition and control, trained on diverse datasets. These models are ideal for generalized humanoid robot reasoning and skills, taking multimodal input—including language and images—to perform manipulation tasks in diverse environments. They are trained on an expansive humanoid dataset consisting of real captured data, synthetic data generated using the components of the Isaac GR00T-Mimic blueprint, Isaac GR00T-Dreams blueprint, and internet-scale video data. These models can easily generalize across common tasks—such as grasping, moving objects with one or both arms, and transferring items from one arm to another—or perform multi-step tasks that require long context and combinations of general skills.
• Simulation Frameworks: Built on NVIDIA Omniverse™ and Cosmos™ for realistic simulations. These frameworks help bridge the simulation-to-reality gap, enabling developers to create and test AI-driven robotics solutions in physically based virtual environments.
• Data Pipelines: Generate synthetic data and environments to train robots. The GR00T-Mimic blueprint allows for the generation of large synthetic motion trajectories from a small number of human demonstrations, while the GR00T-Dreams blueprint generates vast amounts of synthetic trajectory data to teach humanoid robots to perform new actions in novel environments.
• Integrated Computing: Utilizes Jetson AGX Thor™ for running the entire robot stack. This high-performance computing platform is based on NVIDIA Blackwell architecture and comes with integrated functional safety, high-performance CPU, and 100 GB of ethernet bandwidth.
• Modular Workflows: Customizable workflows for various robotic tasks and environments. These include GR00T-Teleop for collecting high-quality human demonstrations via teleoperation, GR00T-Gen for teaching robots to generalize and adapt to new situations using diverse, realistic simulation environments, GR00T-Dexterity for simplifying complex grasping tasks with an end-to-end, pixels-to-action grasping system, GR00T-Mobility for enhancing the robot’s ability to adapt to new, unseen environments using reinforcement learning and imitation learning techniques, GR00T-Control for achieving responsive and precise humanoid robot control with a suite of whole-body control libraries, models, and policies, and GR00T-Perception for integrating vision language models, LLMs, and retrieval-augmented memory to boost perception, cognition, and adaptability.

Benefits

• Accelerated Development: Speeds up the development of humanoid robots by providing a comprehensive platform for training and testing AI-driven robotics solutions.
• High-Fidelity Simulations: Ensures realistic robot interactions with the environment through high-fidelity simulations.
• Versatility: Applicable to a wide range of robotic applications and industries, including material handling, packaging, and inspection.
• Scalability: Supports large-scale training and deployment, making it suitable for complex reinforcement learning environments.
• Enhanced Realism: Provides high-fidelity simulations for realistic robot interactions, ensuring that robots can perform tasks accurately in real-world scenarios.