Skip to content
Robotics

SONIC-DR 2x tracker

whole-body трекер для Unitree G1

2 min readupdated

Это ONNX-политика для гуманоида Unitree G1 29-DoF: каждые 20 мс получает состояние робота и референсное движение (поза таза и 29 углов суставов), а выдаёт 29 целевых позиций для моторных PD-контроллеров. Обучена по стабильному рецепту SuperMimic S3G с SONIC-рандомизацией, сетью на 17,8 млн параметров и look-ahead целью.

Ключевые особенности

  • ONNX opset 17, 71 МБ; нормализаторы observation и goal уже внутри графа.
  • Goal — не следующие 10 кадров, а кадры t+1, t+6, …, t+46: шаг 5, горизонт до 0,92 с. Старый deploy без goal_stride молча подаёт неверную цель.
  • Обучение с hidden-reference fall recovery: при падении достаточно удерживать stand pose, политика встаёт сама.
  • Вариант без резиновых рук; масса торса и запястий рандомизирована x0.8–1.5, трение 0.3–1.6, есть толчки и 10% клипов с замороженными кадрами.
  • В RoboJuDo sim2sim все девять демо-движений прошли на 1.0x и 0.7x без падений.

Пример

import json
import numpy as np
import onnxruntime as ort
name = 's3gla_nohand_sonicdr_2x_ckpt030000'
meta = json.load(open(f'{name}/{name}.json'))
session = ort.InferenceSession(f'{name}/{name}.onnx', providers=['CPUExecutionProvider'])
action = session.run(['action'], {'obs_hist': np.zeros((1, 10, 99), np.float32), 'goal': np.zeros((1, 730), np.float32)})[0]
ПодробнееБенчмарки в MuJoCo: normalized tracking return 0,653 против 0,620 у S3G 30k; failed clips 7/40 против 11/40; на held-out наборах failure rate 0,7%/0,4%, HI-MPJPE 27,6/27,8 мм.
Реальные конфиги идут на CPU Jetson: около 2,2 мс на шаг при 50 Гц; нужен onnxruntime.
На реальном G1 модель ещё не запускали — только sim2sim и офлайн-проверки. Слабые места: высокие удары и подъёмы колена (стопа может оставаться на 25+ см ниже референса), вставание на 1–2 с дольше, чем у S3G.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.1 GB · ready files0.6 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.

    # seu-zuxing/g1-sonicdr-2x-tracker
    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 seu-zuxing/g1-sonicdr-2x-tracker --local-dir ~/models/g1-sonicdr-2x-tracker
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to g1-sonicdr-2x-tracker
  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 g1-sonicdr-2x-tracker (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

seu-zuxing/g1-sonicdr-2x-tracker
Author
seu-zuxing
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Robotics

GR00T N1.7

GR00T N1.7