AB
AB
AiBoss

All Categories

project

GR00T N1 - NVIDIA's open-source basic model for humanoid robots

The GR00T N1 is NVIDIA's first open-source foundational model, designed specifically for general-purpose humanoid robots. It enables maneuvering tasks in diverse environments based on multimodal inputs (such as language and images). The GR00T N1 is based on large-scale...

What is GR00T N1?

GR00T N1 is NVIDIA's first open-source foundational model, designed specifically for general-purpose humanoid robots. It enables maneuvering tasks in diverse environments based on multimodal inputs such as language and images. GR00T N1 is trained on a large-scale humanoid robot dataset, combining real-world, synthetic, and internet video data, and then post-trained to adapt to specific robot forms, tasks, and environments. GR00T N1 employs a dual-system architecture: a vision-language model handles reasoning and planning, while a diffusion transformer generates precise movements. GR00T N1 performs exceptionally well in both simulated and real-world testing, demonstrating significant advantages in complex multi-step tasks and precise manipulation, providing efficient solutions for applications such as material handling, packaging, and inspection.

Main functions of GR00T N1

  • General operation task executionIt can perform various operational tasks in diverse environments, such as grasping, carrying, and coordinated two-arm operation.
  • Multimodal input processingIt processes both language commands and visual images simultaneously, enabling the robot to perform complex tasks based on natural language instructions.
  • Cross-robot morphology adaptabilityIt adapts to different types of robot platforms (such as Fourier GR-1 and 1X Neo) to achieve versatility.
  • Complex Task Reasoning and Planning: Performing complex, multi-step tasks that require continuous contextual understanding and the integration of multiple skills.
  • Efficient data utilization and trainingBy combining internet-scale data, synthetic data, and real robot data for pre-training, performance and generalization ability are significantly improved, and dependence on large-scale labeled data is reduced.

Technical principles of GR00T N1

  • Dual system architecture:
    • Visual-Language Model (System 2)Built on NVIDIA-Eagle and SmolLM-1.7B, it is responsible for understanding the environment using visual and verbal commands, performing reasoning and planning, and outputting action plans.
    • Diffusion converter (System 1)As a motion model, it translates the plans of the vision-language model into precise continuous actions to control the robot's movement.
  • Data StrategyThe pre-training data includes internet video data (providing human action patterns and task semantics), synthetic data (generated on the NVIDIA Omniverse platform, supplementing motion control signals), and real robot data (collected through teleoperation to ensure the model adapts to real-world environments). Unsupervised learning extracts motion patterns from large-scale unlabeled human video data, improving robot learning efficiency.
  • Model training and optimizationPre-training on large-scale data allows the model to learn general motion and operation patterns. Fine-tuning is then performed for specific robot platforms, tasks, and environments to further improve the model's adaptability and performance. During the inference phase, computational efficiency is optimized by reducing diffusion steps to ensure real-time performance.

Project address of GR00T N1

Application scenarios of GR00T N1

  • Logistics and WarehousingUsed for grabbing, moving and sorting goods, automatically taking inventory, and optimizing goods storage and management.
  • manufacturingTo perform precise assembly of parts, conduct product quality inspection, and improve production efficiency and quality control.
  • retail industryAutomatically organizes shelves and replenishes stock, provides customers with information inquiry and product recommendation services, and enhances the shopping experience.
  • healthcare: Assist patients in rehabilitation training, transport and manage medical supplies, and reduce the burden on medical staff.
  • Industrial Inspection and MaintenanceConduct routine inspections of equipment, identify and report any abnormalities; perform simple maintenance operations to reduce labor costs.
Back