Embodied AI Glossary中文GR00T-WholeBodyControl
GR00T 全身控制代码库GR00T WBCAdvancedNVIDIA's open-source codebase for training, evaluating, and deploying whole-body controllers for humanoid robots.
GR00T-WholeBodyControl is a codebase NVIDIA open-sourced on GitHub (under NVlabs), providing model weights for humanoid whole-body controllers along with training, evaluation, and deployment scripts. Whole-body control means coordinating the legs, torso, and arms together, so a robot keeps its balance while walking and manipulating objects at the same time. According to the repository, it includes the decoupled whole-body controller used by GR00T N1.5 and N1.6 on the Unitree G1 (a reinforcement-learning policy handling lower-body walking and balance, with inverse kinematics tracking hand targets for the upper body), as well as controllers like SONIC that are based on large-scale motion tracking. Within the GR00T system, it sits at the “cerebellum” layer: the upper-level VLA supplies a goal, and this layer converts it into joint commands.
ExampleGR00T N1.6 outputs a walking-velocity command and target poses for both hands on a Unitree G1, and the decoupled controller in this codebase converts them into whole-body joint commands.
- Also called
- GR00T WBC, GR00T whole-body control platform
- Related
- Whole-Body Control · Decoupled Whole-Body Control · NVIDIA Isaac GR00T N1 · SONIC · Unitree G1 · NVIDIA Generalist Embodied Agent Research Lab
- Sources
- NVlabs/GR00T-WholeBodyControl (GitHub)
- As of
- 2025-12