GR00T-WholeBodyControl Documentation#
Welcome to the official documentation for GR00T Whole-Body Control (WBC)! This is a unified platform for developing and deploying advanced humanoid controllers.
What is GR00T-WholeBodyControl?#
This codebase serves as the foundation for:
Decoupled WBC models used in NVIDIA Isaac-Gr00t, Gr00t N1.5 and N1.6 (see detailed reference)
GEAR-SONIC Series: State-of-the-art controllers from the GEAR team
News#
[2026-04-14] Live web demo â try SONIC interactively in your browser. Features Kimodo text-to-motion generation.
[2026-04-10] Released SONIC training code and checkpoint on HuggingFace. Train from scratch or finetune. Additional embodiment support and VLA data collection pipeline. See Training Guide.
[2026-03-24] C++ inference stack update: motor error monitoring, TTS alerts, ZMQ protocol v4, idle-mode readaptation. ZMQ header size changed to 1280 bytes.
[2026-03-16] BONES-SEED open-sourced â 142K+ human motions (~288 hours) with G1 MuJoCo trajectories.
[2026-02-19] Released GEAR-SONIC: pretrained checkpoints, C++ inference, VR teleoperation, and documentation.
[2025-11-12] Initial release with Decoupled WBC for GR00T N1.5 and N1.6.
GEAR-SONIC#
SONIC is a humanoid behavior foundation model that gives robots a core set of motor skills learned from large-scale human motion data. Rather than building separate controllers for every motion, SONIC uses motion tracking as a scalable training task so a single unified policy can produce natural, whole-body movement and support a wide range of behaviors.
đ¯ Key Features:
đļ Natural whole-body locomotion (walking, crawling, dynamic movements)
đŽ Real-time VR teleoperation support
đ¤ Foundation for higher-level planning and interaction
đĻ Ready-to-deploy C++ inference stack
Quick Start: Sim2Sim#
Quickly test the SONIC deployment stack in MuJoCo before deploying on real hardware.
Tip
Get running in minutes! Follow the Installation and Quickstart guides to see this in action on your machine.
Documentation#
Getting Started
Tutorials
Best Practices
- Whole-body Teleoperation Guide
- Troubleshooting
- 1.
ModuleNotFoundError: No module named 'isaaclab' - 2. Mesh files are tiny text files (Git LFS not installed)
- 3.
RuntimeError: size mismatchwhen loading a checkpoint - 4.
trl/transformersversion conflict during pip install - 5. TensorRT build fails (
TensorRT_ROOTnot set) - 6. Motion file path errors (
FileNotFoundErroror empty motion library) - 7. Body name errors (
RuntimeError: body 'xxx' not found) - 8. Robot explodes or falls immediately on first frame
- 9. Robot behaves weirdly during deployment (wrong TensorRT version)
- 10.
ChannelFactory create domain errorin MuJoCo sim - 11. SMPL tracking is unstable or drifts
- 12. MuJoCo viewer renders incorrectly in Docker
- Still stuck?
- 1.
Reference Documentation
- Reference Documentation
- Configuration Guide
- Coordinate Frame and Rotation Conventions
- Training Code Structure
- C++ Deployment Program Flow
- Observation Configuration
- Motion Reference Data
- Folder Structure
- The C++ reader scans the base directory for subfolders, reads each subfolder as one motion, and validates frame-count consistency across all files in that folder.
- File Formats
- Creating Your Own Reference Motions
- Verifying Reference Motions
- Using Reference Motions
- Validation Rules
- Notes
- Kinematic Planner ONNX Model Reference
- G1 JetPack 6 Flashing Guide
- Decoupled WBC