data_folder / README.md
RoboDojo Asset Ops
Describe the current dataset concisely
7d4ce8b
|
Raw History Blame Contribute Delete
2.85 kB
metadata
language:
  - zh
  - en
pretty_name: data_folder
tags:
  - robotics
  - simulation
  - robodojo
configs:
  - config_name: native_episode_inventory
    data_files:
      - split: inventory
        path: native_episode_inventory.csv
extra_gated_heading: 申请访问训练数据 / Request training data access
extra_gated_description: >-
  申请需经人工审核,获批后方可下载训练数据。所在地及单位信息用于申请审核与地域、机构统计,联系邮箱用于申请相关联系,无需提供街道或详细地址。Requests
  are reviewed manually. Location and organization information is used for
  access review and geographic/organization statistics; the contact email is
  used for application-related communication. No street address is required.
extra_gated_prompt: >-
  请填写当前主要工作或研究所在地,而非国籍。示例:中国 / 广东省 / 深圳市。没有省或州的地区填写 N/A。Please enter your
  current primary work or research location, not your nationality. Example:
  China / Guangdong / Shenzhen. Enter N/A if no province or state applies.
  请填写单位全称(学校、研究机构或公司;无所属单位请填“独立研究者”),并提供可接收邮件的联系邮箱。Please provide your full
  organization name (university, research institute, or company; enter
  "Independent researcher" if unaffiliated) and a reachable contact email.
extra_gated_button_content: 提交申请 / Submit request
extra_gated_fields:
  国家/地区 / Country or region: country
  省/州 / Province or state: text
  城市 / City: text
  单位 / Organization: text
  联系邮箱 / Contact email: text

RoboDojo 训练数据

共 10,800 条,机器人为 franka、piper、x5,每个机器人、每个任务各 1,200 条。

任务 数据量
general_pickup 3,600
place_the_cup_on_the_table 3,600
place_the_goods_into_basket 3,600

general_pickup 包含标准场景和干扰物场景,以 task_variant 区分。目录按 <task>/<robot>/ 组织,包含 HDF5、五路 MP4、布局、场景和种子信息。

元文件 用途
native_episode_inventory.csv 逐条索引,包含任务、机器人、编号、帧数和来源哈希。
selection_manifest.json 逐条来源、原始编号、质量信息和文件哈希。
publication_summary.json 数据统计、质量标准与验证范围摘要。
document_checksums.json 说明和索引文件的哈希校验。
task_merge_manifest.json 仅本地归档的合并映射;公开对应信息已在数据清单及场景文件中保留。

仓库公开可见,下载训练数据需申请并经人工审核。