Skip to content
Robotics

pi05 для mobile_manipulator

бросает банку-томат в мусорку

2 min read

Это VLA-политика (vision-language-action: модель получает изображения с камер и состояние робота, а выдаёт команды действия) на базе lerobot/pi05_base от Physical Intelligence. Её дообучили в LeRobot под мобильный манипулятор: робот с камерой на голове и левой руке должен бросить томатную банку в мусорное ведро. На входе — состояние из 20 значений и два изображения, на выходе — 20-мерный вектор action. Это узкий файнтюн под одну задачу, а не универсальный робот; он пригодится, чтобы воспроизвести пайплайн LeRobot и потренироваться на похожих манипуляциях.

Ключевые особенности

  • База — lerobot/pi05_base, дообучение под mobile_manipulator.
  • Две камеры: head_camera 480×640 и hand_left_camera 400×640.
  • Одна задача: «Throw the tomato can into the trash bin».
  • Датасет: 200 эпизодов, 48 141 кадр, 10 FPS.
  • Оценки успешности на роботе пока нет.

Что можно сделать

  • Запустить политику через lerobot-rollout на своём mobile_manipulator и проверить поведение.
  • Собрать похожий датасет и дообучить pi05_base под свою задачу манипуляции.

Пример

lerobot-rollout \
  --strategy.type=base \
  --robot.type=mobile_manipulator \
  --robot.port=<your_robot_port> \
  --policy.path=team-sobits/sobit_home_left_sim-pnp_tomato_trash-abs-200ep-pi05_vlmlora_nostate-steps30k-decay30k \
  --task="Throw the tomato can into the trash bin" \
  --duration=60
ПодробнееДообучение: 30 000 шагов, batch size 32, AdamW, lr 2.5e-05.
Данные: 200 эпизодов, 48 141 кадр, 10 FPS.
Вход: state (20,) + два изображения; выход: action (20,).
Оценки успешности пока нет; лицензия Apache-2.0.
Discuss Telegram 0

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 14 GB · ready files16 GB
  • BF16weights 6.9 GB8.1 GB
  • FP8 / INT8recommendedweights 3.6 GB4.4 GB
  • 4-bitweights 2.1 GB2.8 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 · F32 — from RTX 5080

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

This is a LoRA: counted together with its base model pi05_base.

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.

    # team-sobits/sobit_home_left_sim-pnp_tomato_trash-abs-200ep-pi05_vlmlora_nostate-steps30k-decay30k
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 lerobot/pi05_base --local-dir ~/models/pi05_base
    hf download team-sobits/sobit_home_left_sim-pnp_tomato_trash-abs-200ep-pi05_vlmlora_nostate-steps30k-decay30k --local-dir ~/models/sobit_home_left_sim-pnp_tomato_trash-abs-200ep-pi05_vlmlora_nostate-steps30k-decay30k
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to pi05_base
  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

Requires the base model

lerobot/pi05_base

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. paligemma-3b-pt-224family root
  2. pi05_base
  3. sobit_home_left_sim-pnp_tomato_trash-abs-200ep-pi05_vlmlora_nostate-steps30k-decay30k

New posts about versions of paligemma-3b-pt-224 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

team-sobits/sobit_home_left_sim-pnp_tomato_trash-abs-200ep-pi05_vlmlora_nostate-steps30k-decay30k
Author
team-sobits
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