hub-recording

Use for data recording nodes in dora. Triggers on: llama-factory-recorder, lerobot-dashboard, lerobot, recording, data collection, imitation learning, demonstration, training data, dataset, 数据记录, 模仿学习, 演示数据, 训练数据

跳到安装

仓库: ZhangHanDong/dora-skills
最近来源活动: 2026年1月21日 16:16
检测到的 SKILL.md 语言: 英语
星标: 7
分支: 1

安装方式

默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

直接命令不会经过审查 Prompt；运行前请先检查来源。

npx skills add https://github.com/ZhangHanDong/dora-skills --skill hub-recording

命令会保持在同一行。复制前请横向滚动并检查完整内容。

想先保存到本地？可下载 SkillsMP 当前能够提供的文件。

检查来源文件

决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。

SKILL.md

来源说明 · 只读预览

返回搜索返回安装

cli-commands

Use for dora CLI command questions. Triggers on: dora run, dora build, dora start, dora stop, dora new, dora up, dora destroy, dora list, dora ps, dora logs, dora check, dora graph, dora daemon, dora coordinator, CLI命令, 命令行, 运行数据流, 构建, 启动, 停止

2026-01-21

data-pipeline

Use for data recording and replay questions with dora. Triggers on: recording, replay, lerobot, data collection, dataset, rosbag, record data, play back, training data, HDF5, parquet, 数据记录, 数据回放, 数据收集, 训练数据

2026-01-21

dataflow-config

Use for dora-rs dataflow YAML configuration questions. Triggers on: dataflow.yml, dataflow.yaml, nodes:, inputs:, outputs:, timer, dora/timer, queue_size, env:, build:, path:, git:, branch:, tag:, restart_policy, send_stdout_as, operator:, operators:, YAML配置, 数据流配置, 节点配置, 输入输出, 定时器

2026-01-21

查看这个仓库的全部 Skills

name	hub-recording
description	Use for data recording nodes in dora. Triggers on: llama-factory-recorder, lerobot-dashboard, lerobot, recording, data collection, imitation learning, demonstration, training data, dataset, 数据记录, 模仿学习, 演示数据, 训练数据
globs	["/dataflow.yml","/dataflow.yaml"]
source	https://github.com/dora-rs/dora-hub

Data Recording Nodes

Record demonstrations for imitation learning and model fine-tuning

Available Recording Nodes

Node	Install	Description
llama-factory-recorder	`pip install llama-factory-recorder`	Record for LLM/VLM fine-tuning
lerobot-dashboard	`pip install lerobot-dashboard`	LeRobot recording interface
dora-lerobot-recorder	From dora-lerobot repo	LeRobot data recording
dora-rdt-1b	`pip install dora-rdt-1b`	VLA policy inference

dora-lerobot Installation

dora-lerobot is from a separate repository, not PyPI:

git clone https://github.com/dora-rs/dora-lerobot
cd dora-lerobot
pip install -e dora_lerobot

llama-factory-recorder

Record data for fine-tuning language and vision-language models with LLaMA Factory.

Prerequisites

git clone https://github.com/hiyouga/LLaMA-Factory --depth 1 $HOME/LLaMA-Factory

YAML Configuration

- id: recorder
  build: pip install llama-factory-recorder
  path: llama-factory-recorder
  inputs:
    image_right: camera/image
    ground_truth: human/text  # Human-provided labels/responses
  outputs:
    - text
  env:
    DEFAULT_QUESTION: "Respond to people."
    LLAMA_FACTORY_ROOT_PATH: $HOME/LLaMA-Factory

Recording Pipeline

nodes:
  # Camera
  - id: camera
    build: pip install opencv-video-capture
    path: opencv-video-capture
    inputs:
      tick: dora/timer/millis/100
    outputs:
      - image

  # Keyboard input for labels
  - id: keyboard
    build: pip install terminal-input
    path: terminal-input
    outputs:
      - text

  # Recorder
  - id: recorder
    build: pip install llama-factory-recorder
    path: llama-factory-recorder
    inputs:
      image_right: camera/image
      ground_truth: keyboard/text
    outputs:
      - text
    env:
      DEFAULT_QUESTION: "What action should the robot take?"
      LLAMA_FACTORY_ROOT_PATH: $HOME/LLaMA-Factory

  # Visualization
  - id: rerun
    build: pip install dora-rerun
    path: dora-rerun
    inputs:
      image: camera/image

Training with Recorded Data

Run your dataflow to collect data
Data is saved to LLaMA Factory folder
Modify training config:

# llama-factory/examples/train_lora/qwen2vl_lora_sft.yaml
dataset: dora_demo_1,identity  # Your recorded dataset
model_name_or_path: Qwen/Qwen2.5-VL-3B-Instruct  # or 7B

Train:

llamafactory-cli train examples/train_lora/qwen2vl_lora_sft.yaml

lerobot-dashboard

Pygame-based interface for LeRobot data collection with dual camera display.

YAML Configuration

- id: dashboard
  build: pip install lerobot-dashboard
  path: lerobot-dashboard
  inputs:
    tick: dora/timer/millis/16  # 60fps update
    image_left: camera_left/image
    image_right: camera_right/image
  outputs:
    - text      # User text input
    - episode   # Episode number (-1 marks end)
    - failed    # Failed episode number
    - end       # End signal for dataflow
  env:
    WINDOW_WIDTH: 1280
    WINDOW_HEIGHT: 1080

Outputs

Output	Description
text	StringArray - user text input
episode	Int - current episode number (-1 = episode end)
failed	Int - marks episode as failed
end	Empty array - signals recording end

LeRobot Recording Pipeline

nodes:
  # Left camera
  - id: camera_left
    build: pip install opencv-video-capture
    path: opencv-video-capture
    inputs:
      tick: dora/timer/millis/33
    outputs:
      - image
    env:
      PATH: "0"

  # Right camera
  - id: camera_right
    build: pip install opencv-video-capture
    path: opencv-video-capture
    inputs:
      tick: dora/timer/millis/33
    outputs:
      - image
    env:
      PATH: "1"

  # Leader arm (teleoperation)
  - id: leader
    build: pip install dora-piper
    path: dora-piper
    inputs:
      tick: dora/timer/millis/20
    outputs:
      - joint_state
    env:
      MODE: leader

  # Follower arm
  - id: follower
    build: pip install dora-piper
    path: dora-piper
    inputs:
      joint_positions: leader/joint_state
    outputs:
      - joint_state
    env:
      MODE: follower

  # Dashboard
  - id: dashboard
    build: pip install lerobot-dashboard
    path: lerobot-dashboard
    inputs:
      tick: dora/timer/millis/16
      image_left: camera_left/image
      image_right: camera_right/image
    outputs:
      - text
      - episode
      - failed
      - end
    env:
      WINDOW_WIDTH: 1280
      WINDOW_HEIGHT: 720

  # LeRobot recorder (install from dora-lerobot repo first)
  - id: lerobot
    path: dora-lerobot-recorder
    inputs:
      image_left: camera_left/image
      image_right: camera_right/image
      state: follower/joint_state
      action: leader/joint_state
      episode: dashboard/episode
      end: dashboard/end
    env:
      DATASET_NAME: my_robot_dataset

Complete Imitation Learning Pipeline

1. Data Collection

# dataflow_record.yml
nodes:
  - id: camera
    build: pip install opencv-video-capture
    path: opencv-video-capture
    inputs:
      tick: dora/timer/millis/33
    outputs:
      - image

  - id: leader
    build: pip install dora-piper
    path: dora-piper
    inputs:
      tick: dora/timer/millis/20
    outputs:
      - joint_state
    env:
      MODE: leader

  - id: follower
    build: pip install dora-piper
    path: dora-piper
    inputs:
      joint_positions: leader/joint_state
    outputs:
      - joint_state
    env:
      MODE: follower

  - id: dashboard
    build: pip install lerobot-dashboard
    path: lerobot-dashboard
    inputs:
      tick: dora/timer/millis/16
      image_left: camera/image
    outputs:
      - episode
      - end

  # LeRobot recorder (install from dora-lerobot repo first)
  - id: recorder
    path: dora-lerobot-recorder
    inputs:
      image: camera/image
      state: follower/joint_state
      action: leader/joint_state
      episode: dashboard/episode
    env:
      DATASET_NAME: pick_and_place

2. Policy Training (offline)

# Train with LeRobot
python lerobot/train.py \
    --dataset pick_and_place \
    --policy diffusion

3. Policy Deployment

# dataflow_deploy.yml
nodes:
  - id: camera
    build: pip install opencv-video-capture
    path: opencv-video-capture
    inputs:
      tick: dora/timer/millis/33
    outputs:
      - image

  - id: policy
    build: pip install dora-rdt-1b
    path: dora-rdt-1b
    inputs:
      image: camera/image
    outputs:
      - action

  - id: robot
    build: pip install dora-piper
    path: dora-piper
    inputs:
      joint_positions: policy/action
    outputs:
      - joint_state

Data Formats

Episode Markers

# Start new episode
node.send_output("episode", pa.array([episode_number]))

# End episode
node.send_output("episode", pa.array([-1]))

# Mark episode as failed
node.send_output("failed", pa.array([episode_number]))

Recording State/Action Pairs

# State: current robot joint positions
state = pa.array(current_joints)

# Action: commanded joint positions (from leader/policy)
action = pa.array(target_joints)

Related Skills

hub-robot - Robot control for teleoperation
hub-camera - Camera input for recording
data-pipeline - LeRobot data tools

hub-recording

来源信息

安装方式

检查来源文件

同仓库更多 Skills

Data Recording Nodes

Available Recording Nodes

dora-lerobot Installation

llama-factory-recorder

Prerequisites

YAML Configuration

Recording Pipeline

Training with Recorded Data

lerobot-dashboard

YAML Configuration

Outputs

LeRobot Recording Pipeline

Complete Imitation Learning Pipeline

1. Data Collection

2. Policy Training (offline)

3. Policy Deployment

Data Formats

Episode Markers

Recording State/Action Pairs

Related Skills

同仓库更多 Skills

Data Recording Nodes

Available Recording Nodes

dora-lerobot Installation

llama-factory-recorder

Prerequisites

YAML Configuration

Recording Pipeline

Training with Recorded Data

lerobot-dashboard

YAML Configuration

Outputs

LeRobot Recording Pipeline

Complete Imitation Learning Pipeline

1. Data Collection

2. Policy Training (offline)

3. Policy Deployment

Data Formats

Episode Markers

Recording State/Action Pairs

Related Skills