Contribute Dataset贡献数据集

1 · Hugging Face data format1 · Hugging Face 数据格式

Keep the original LeRobot dataset generated by BREE recording. The current recorder uses v3.0; retain the format version in meta/info.json. Document other versions explicitly.保留 BREE 采集生成的原始 LeRobot 数据集。当前采集实现使用 v3.0；保留 meta/info.json 中的格式版本，其他版本需明确注明。

dataset/
├── meta/info.json
├── meta/stats.json
├── meta/tasks.parquet
├── meta/episodes/chunk-000/file-000.parquet
├── data/chunk-000/file-000.parquet
└── videos/{video_key}/chunk-000/file-000.mp4

Illustrative v3.0 paths. Preserve the generated chunking and the paths declared in the metadata.以上为 v3.0 路径示意；保留实际生成的分块和元数据中声明的路径。

• State and action: observation.state and action, each with 12 dimensions for the current dual-arm setup. Preserve joint names, ordering and normalization settings.状态与动作：当前双臂配置的 observation.state 和 action 均为 12 维。保留关节名称、顺序和归一化设置。
• Three camera streams: observation.images.left_global, observation.images.left_wrist and observation.images.right_wrist. The recording template uses 640 × 480 at 30 fps.三路相机：observation.images.left_global、observation.images.left_wrist 和 observation.images.right_wrist。采集模板采用 640 × 480、30 fps。
• Preserve episode, frame, timestamp and task indices. Include task descriptions and synchronized videos; document missing frames or incomplete episodes.保留回合、帧、时间戳与任务索引，提供任务描述和同步视频；注明缺帧或未完成的回合。

2 · Robot and camera calibration2 · 机器人与相机校准

• Calibrate all four arms with the physical midpoint fixture for first setup or after relevant hardware changes. Check ports and calibration IDs. Routine startup does not require rewriting calibration.首次搭建或相关硬件变更后，使用物理中点夹具完成四条机械臂校准，并核对端口与标定 ID。日常开机无需重复写入标定。
• Before recording, verify the follower-arm H poses and leader–follower alignment.采集前检查双从臂 H 位，以及主从臂对齐状态。
• Align global and both wrist views with saved references. Keep camera mounts, imaging settings, lighting and background consistent.将全局和左右腕部画面与保存的参考图对齐，保持相机安装、成像参数、光照和背景一致。
• Document the hardware configuration, calibration date and reference images with the dataset. Describe any deviations from the standard setup.随数据集说明记录硬件配置、校准日期与参考图，并注明与标准配置的差异。

3 · Dataset naming and description3 · 数据集命名与说明

A Hugging Face dataset ID has the form <namespace>/<dataset_name>. For new contributions, we suggest a readable name containing the task, anonymous setup identifier and collection version:Hugging Face 数据集 ID 使用 <namespace>/<dataset_name>。新贡献建议在名称中包含任务、匿名设备编号和采集版本：

<namespace>/bree_task01_pickplace_site01_v1

This is a suggested naming example. Use the task reference to identify the task; document the mapping when retaining a historical dataset name.以上为建议命名示例。任务编号以任务参考页为准；保留历史数据集名称时，请注明任务对应关系。

Include a dataset README with the task and instruction, episode count, frame rate, camera keys, format version, collection conditions, calibration record and data license. During anonymous review, use anonymous identifiers in descriptions and links.数据集 README 应说明任务与指令、回合数、帧率、相机键、格式版本、采集条件、校准记录和数据许可。匿名审稿期间，说明与链接使用匿名标识。

4 · Planned PR review workflow4 · 后续 PR 审核流程

After anonymous review, this page will link to the formal community contribution repository. The planned workflow is:匿名审稿结束后，本页将接入正式的社区数据贡献仓库，拟采用以下流程：

1. Prepare the dataset, dataset README and calibration documentation.整理数据集、数据集 README 与校准记录。
2. Publish the dataset to Hugging Face, then open a PR with its dataset ID, revision and contribution summary.将数据发布至 Hugging Face，再通过 PR 提交数据集 ID、版本与贡献摘要。
3. Review format compatibility, video and episode integrity, calibration documentation and task mapping; request corrections where needed.检查格式兼容性、视频与回合完整性、校准说明和任务对应关系，并反馈需补充或修正的内容。
4. Merge the reviewed dataset entry into the community registry.审核通过后，将数据集条目合并至社区目录。