five

ZhaoRunyi/Piper_click_bell_0403

收藏
Hugging Face2026-04-20 更新2026-04-12 收录
下载链接:
https://hf-mirror.com/datasets/ZhaoRunyi/Piper_click_bell_0403
下载链接
链接失效反馈
官方服务:
资源简介:
# Piper LeRobot Dataset Generated on `2026-04-20` by `convert_piper_hdf5_to_lerobot.py`. ## Summary - Repo id: `ZhaoRunyi/Piper_click_bell_0403` - Robot type: `aloha` - Input dir: `/home/edemlab/challenge_ws/control_your_robot/example/teleop/save/dual_piper_data/click_bell` - Output dir: `/home/edemlab/challenge_ws/data/ZhaoRunyi/Piper_click_bell_0403` - FPS: `10` - Sampling mode: `full` - Leading group: `slave_cam_high` - Crop strategy: `LATEST_START` - Default leading tolerance: `0.1` s - Additional constraints: `[]` - Arm selection: `dual` -> `left, right` - Space ids: `all` -> `['joint', 'gripper', 'ee_pos', 'ee_rot']` - EE rotation format: `rot6d` - Gripper format: `raw` - Image preset: `all` -> `['observation.images.cam_high', 'observation.images.cam_left_wrist', 'observation.images.cam_right_wrist']` ## State / Action Index Map `action` uses the exact same index order as `observation.state`. | indices | arm | component | names | source | | --- | --- | --- | --- | --- | | 0-5 | left | joint | left_joint_waist, left_joint_shoulder, left_joint_elbow, left_joint_forearm_roll, left_joint_wrist_angle, left_joint_wrist_rotate | slave_left_arm/joint | | 6 | left | gripper | left_gripper | slave_left_arm/gripper | | 7-9 | left | ee_pos | left_ee_pos_x, left_ee_pos_y, left_ee_pos_z | slave_left_arm/ee_pose[:3] | | 10-15 | left | ee_rot | left_ee_rot6d_0, left_ee_rot6d_1, left_ee_rot6d_2, left_ee_rot6d_3, left_ee_rot6d_4, left_ee_rot6d_5 | slave_left_arm/ee_pose[3:6] (controller-scaled Piper xyz Euler) -> rot6d | | 16-21 | right | joint | right_joint_waist, right_joint_shoulder, right_joint_elbow, right_joint_forearm_roll, right_joint_wrist_angle, right_joint_wrist_rotate | slave_right_arm/joint | | 22 | right | gripper | right_gripper | slave_right_arm/gripper | | 23-25 | right | ee_pos | right_ee_pos_x, right_ee_pos_y, right_ee_pos_z | slave_right_arm/ee_pose[:3] | | 26-31 | right | ee_rot | right_ee_rot6d_0, right_ee_rot6d_1, right_ee_rot6d_2, right_ee_rot6d_3, right_ee_rot6d_4, right_ee_rot6d_5 | slave_right_arm/ee_pose[3:6] (controller-scaled Piper xyz Euler) -> rot6d | ## Notes - Timestamp alignment is done per HDF5 group. Each camera group and each arm group has its own `timestamp` dataset, which is treated like an individual topic timeline. - The leading group is `slave_cam_high`. In `full` mode every leading-group timestep is kept; in `fps` mode the leading timeline is resampled first and then all other groups are bound to it by nearest timestamp. - Piper HDF5 stores 6 arm joints plus one separate gripper scalar per arm. When both `joint` and `gripper` are enabled, the gripper is appended as the 7th dimension of that arm's joint block. - `ee_pose` rotation is decoded from the current PiperController convention: controller value -> multiply by 1000 -> Piper xyz Euler degrees -> convert to `rot6d`. When `EE_ROT_FORMAT` is `rpy`/`euler`, the exported values are radians. - `GRIPPER_FORMAT="01"` treats values in `[0, 1]` as normalized opening ratio and thresholds them with `GRIPPER_FULL_WIDTH=0.05` and `GRIPPER_THRESHOLD=0.01`. ## Image Keys | dataset key | hdf5 group | | --- | --- | | observation.images.cam_high | slave_cam_high | | observation.images.cam_left_wrist | slave_cam_left | | observation.images.cam_right_wrist | slave_cam_right | ## Feature Names `observation.state` / `action` names: ```text left_joint_waist, left_joint_shoulder, left_joint_elbow, left_joint_forearm_roll, left_joint_wrist_angle, left_joint_wrist_rotate, left_gripper, left_ee_pos_x, left_ee_pos_y, left_ee_pos_z, left_ee_rot6d_0, left_ee_rot6d_1, left_ee_rot6d_2, left_ee_rot6d_3, left_ee_rot6d_4, left_ee_rot6d_5, right_joint_waist, right_joint_shoulder, right_joint_elbow, right_joint_forearm_roll, right_joint_wrist_angle, right_joint_wrist_rotate, right_gripper, right_ee_pos_x, right_ee_pos_y, right_ee_pos_z, right_ee_rot6d_0, right_ee_rot6d_1, right_ee_rot6d_2, right_ee_rot6d_3, right_ee_rot6d_4, right_ee_rot6d_5 ```

# Piper LeRobot数据集 本数据集由`convert_piper_hdf5_to_lerobot.py`于`2026-04-20`生成。 ## 摘要 - 仓库ID:`ZhaoRunyi/Piper_click_bell_0403` - 机器人类型:`aloha` - 输入目录:`/home/edemlab/challenge_ws/control_your_robot/example/teleop/save/dual_piper_data/click_bell` - 输出目录:`/home/edemlab/challenge_ws/data/ZhaoRunyi/Piper_click_bell_0403` - 帧率(FPS):`10` - 采样模式:`full` - 主导组:`slave_cam_high` - 裁剪策略:`LATEST_START` - 默认主导容差:`0.1` s - 附加约束:`[]` - 机械臂选择:`dual` -> `左、右` - 空间维度标识:`all` -> `['joint', 'gripper', 'ee_pos', 'ee_rot']` - 末端执行器(EE)旋转格式:`rot6d` - 夹爪(gripper)格式:`raw` - 图像预设:`all` -> `['observation.images.cam_high', 'observation.images.cam_left_wrist', 'observation.images.cam_right_wrist']` ## 状态/动作索引映射 `action`与`observation.state`使用完全一致的索引顺序。 | 索引范围 | 机械臂 | 组件类型 | 名称 | 数据源 | | --- | --- | --- | --- | --- | | 0-5 | 左 | 关节 | left_joint_waist, left_joint_shoulder, left_joint_elbow, left_joint_forearm_roll, left_joint_wrist_angle, left_joint_wrist_rotate | slave_left_arm/joint | | 6 | 左 | 夹爪 | left_gripper | slave_left_arm/gripper | | 7-9 | 左 | 末端执行器位置(EE pos) | left_ee_pos_x, left_ee_pos_y, left_ee_pos_z | slave_left_arm/ee_pose[:3] | | 10-15 | 左 | 末端执行器旋转(EE rot) | left_ee_rot6d_0, left_ee_rot6d_1, left_ee_rot6d_2, left_ee_rot6d_3, left_ee_rot6d_4, left_ee_rot6d_5 | slave_left_arm/ee_pose[3:6] (controller-scaled Piper xyz Euler) -> rot6d | | 16-21 | 右 | 关节 | right_joint_waist, right_joint_shoulder, right_joint_elbow, right_joint_forearm_roll, right_joint_wrist_angle, right_joint_wrist_rotate | slave_right_arm/joint | | 22 | 右 | 夹爪 | right_gripper | slave_right_arm/gripper | | 23-25 | 右 | 末端执行器位置(EE pos) | right_ee_pos_x, right_ee_pos_y, right_ee_pos_z | slave_right_arm/ee_pose[:3] | | 26-31 | 右 | 末端执行器旋转(EE rot) | right_ee_rot6d_0, right_ee_rot6d_1, right_ee_rot6d_2, right_ee_rot6d_3, right_ee_rot6d_4, right_ee_rot6d_5 | slave_right_arm/ee_pose[3:6] (controller-scaled Piper xyz Euler) -> rot6d | ## 说明 - 时间戳对齐基于HDF5数据组完成。每个相机数据组与每个机械臂数据组均拥有独立的`timestamp`数据集,可视为单独的话题时间线。 - 主导组为`slave_cam_high`。在`full`(全量)模式下,所有主导组时间步均被保留;在`fps`模式下,将首先对主导时间线进行重采样,随后通过最近邻时间戳将其余所有数据组绑定至该时间线。 - Piper格式的HDF5文件为每台机械臂存储6个关节数据与1个独立的夹爪标量数据。当同时启用`joint`(关节)与`gripper`(夹爪)维度时,夹爪数据将作为该机械臂关节块的第7维被追加。 - `ee_pose`(末端执行器位姿)旋转数据基于当前Piper控制器约定解码:控制器数值 -> 乘以1000 -> Piper XYZ欧拉角(角度制) -> 转换为`rot6d`(6维旋转表示)。当`EE_ROT_FORMAT`设置为`rpy`/`euler`时,导出值将以弧度为单位。 - 当`GRIPPER_FORMAT="01"`时,将`[0,1]`区间内的数值视为归一化的夹爪张开比率,并通过`GRIPPER_FULL_WIDTH=0.05`与`GRIPPER_THRESHOLD=0.01`对其进行阈值处理。 ## 图像键值 | 数据集键 | HDF5数据组 | | --- | --- | | observation.images.cam_high | slave_cam_high | | observation.images.cam_left_wrist | slave_cam_left | | observation.images.cam_right_wrist | slave_cam_right | ## 特征名称 `observation.state`与`action`的特征名称如下: text left_joint_waist, left_joint_shoulder, left_joint_elbow, left_joint_forearm_roll, left_joint_wrist_angle, left_joint_wrist_rotate, left_gripper, left_ee_pos_x, left_ee_pos_y, left_ee_pos_z, left_ee_rot6d_0, left_ee_rot6d_1, left_ee_rot6d_2, left_ee_rot6d_3, left_ee_rot6d_4, left_ee_rot6d_5, right_joint_waist, right_joint_shoulder, right_joint_elbow, right_joint_forearm_roll, right_joint_wrist_angle, right_joint_wrist_rotate, right_gripper, right_ee_pos_x, right_ee_pos_y, right_ee_pos_z, right_ee_rot6d_0, right_ee_rot6d_1, right_ee_rot6d_2, right_ee_rot6d_3, right_ee_rot6d_4, right_ee_rot6d_5
提供机构:
ZhaoRunyi
5,000+
优质数据集
54 个
任务类型
进入经典数据集
二维码
社区交流群

面向社区/商业的数据集话题

二维码
科研交流群

面向高校/科研机构的开源数据集话题

数据驱动未来

携手共赢发展

商业合作