Skip to content
Robotics

SONIC residual G1

correction policy for the Unitree G1 robot

1 min readupdated

This model is a residual policy for the 23-DOF humanoid robot Unitree G1, running on top of the frozen base model NVIDIA GEAR-SONIC. The solution corrects the actions of the base decoder on real joints, improving movement accuracy and stability. The model is trained on BONES-SEED motions and is suitable for robotics researchers and developers for testing in the MuJoCo simulator and on real hardware.

Key features

  • Correction of base SONIC model actions for 23 Unitree G1 joints.
  • Trained on the BONES-SEED motion dataset.
  • Noticeable reduction in positioning errors and fall counts in simulation.
  • MLP architecture with observation normalization.

What you can do

  • Build a deployable policy package using the export script for the simulator or a real robot.
  • Evaluate robot robustness against external pushes at speeds up to 2 m/s.
  • Run the MuJoCo simulation using the C++ deployment program.

Example

Exporting the deployment policy from the repository root directory:

bash gear_sonic/scripts/export_deploy_policy.sh residual residual23
cd gear_sonic_deploy
bash deploy.sh --dof-profile g1_23dof --cp policy/residual23/model \
  --obs-config policy/residual23/observation_config.yaml sim
Read moreDetails
Base model: nvidia/GEAR-SONIC
Nominal root position error: 49.1 mm versus 58.4 mm for the base model
Falls at 1 m/s pushes: 2 versus 23 for the base
License: NVIDIA Open Model License
Discuss Telegram 0 25

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # yolkarian/sonic-g1-23dof-residual
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download yolkarian/sonic-g1-23dof-residual --local-dir ~/models/sonic-g1-23dof-residual
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of GEAR-SONIC (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 3All on Hugging Face
Fine-tunes 1All on Hugging Face

About the model

yolkarian/sonic-g1-23dof-residual
Author
yolkarian
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

250

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments