跨机型 HDF5 数据布局参考¶
来源：
ubt_IL/scripts/convert/common/all_robot_h5_info_v1.2.md。涵盖各来源机型的 HDF5 数据（end_effector / joint_position / 相机 / 深度）维度与单位定义，编写自定义转换配置前请先对照本文。
Unit Description¶
The units for the xyz coordinates in the end_effector are: meters (m).
The units for the rpy (roll, pitch, yaw) angles in the end_effector are: radians (rad).
The units for the link angles in joint_position are: radians (rad).
The units for depth_images are: millimeters (mm).
Note¶
When using depth images, pay attention to the units. If the maximum value in the depth image is greater than 100, the units are millimeters (mm); otherwise, the units are meters (m).
It is recommended to include the following check when using depth images:
AgileX 3RGB¶
Note¶
The h5 format is identical to that of the AgileX 3RGB described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
end_effector (7d): [x, y, z, quaternions (x,y,z,w)]
joint_position (7d): [base link, ..., end_effector link, gripper]
Found HDF5 file: ./benchmark1_1_release/h5_agilex_3rgb/20_takecorn_2/success_episodes/train/2024_10_08-16_13_09-172910045527931776.00/data/trajectory.hdf5
File structure:
master (Group)
end_effector_left (Dataset)
Shape: (539, 7), Type: float64
end_effector_right (Dataset)
Shape: (539, 7), Type: float64
joint_effort_left (Dataset)
Shape: (539, 7), Type: float64
joint_effort_right (Dataset)
Shape: (539, 7), Type: float64
joint_position_left (Dataset)
Shape: (539, 7), Type: float64
joint_position_right (Dataset)
Shape: (539, 7), Type: float64
joint_velocity_left (Dataset)
Shape: (539, 7), Type: float64
joint_velocity_right (Dataset)
Shape: (539, 7), Type: float64
observations (Group)
depth_images (Group)
camera_front (Dataset)
Shape: (539,), Type: object
camera_left_wrist (Dataset)
Shape: (539,), Type: object
camera_right_wrist (Dataset)
Shape: (539,), Type: object
rgb_images (Group)
camera_front (Dataset)
Shape: (539,), Type: object
camera_left_wrist (Dataset)
Shape: (539,), Type: object
camera_right_wrist (Dataset)
Shape: (539,), Type: object
puppet (Group)
end_effector_left (Dataset)
Shape: (539, 7), Type: float64
end_effector_right (Dataset)
Shape: (539, 7), Type: float64
joint_effort_left (Dataset)
Shape: (539, 7), Type: float64
joint_effort_right (Dataset)
Shape: (539, 7), Type: float64
joint_position_left (Dataset)
Shape: (539, 7), Type: float64
joint_position_right (Dataset)
Shape: (539, 7), Type: float64
joint_velocity_left (Dataset)
Shape: (539, 7), Type: float64
joint_velocity_right (Dataset)
Shape: (539, 7), Type: float64
Franka 3RGB¶
Note¶
The h5 format is identical to that of the Franka 3RGB described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
end_effector (6d): [x, y, z, r, p, y]
joint_position (8d): [base link, ..., end_effector link, gripper]
Found HDF5 file: ./benchmark1_1_release/h5_franka_3rgb/apples_placed_on_a_ceramic_plate/success_episodes/train/1112_141543/data/trajectory.hdf5
File structure:
master (Group)
joint_position (Dataset)
Shape: (98, 8), Type: float64
observations (Group)
depth_images (Group)
camera_left (Dataset)
Shape: (98,), Type: object
camera_right (Dataset)
Shape: (98,), Type: object
camera_top (Dataset)
Shape: (98,), Type: object
rgb_images (Group)
camera_left (Dataset)
Shape: (98,), Type: object
camera_right (Dataset)
Shape: (98,), Type: object
camera_top (Dataset)
Shape: (98,), Type: object
puppet (Group)
end_effector (Dataset)
Shape: (98, 6), Type: float64
joint_position (Dataset)
Shape: (98, 8), Type: float64
Franka Fr3 Dual Arm¶
end_effector (12d): [end_effector_left, end_effector_right]
end_effector_left/right (6d): [x, y, z, r, p, y]
joint_position (8d): [joint_position_left, joint_position_right]
joint_position_left/right (16d): [base link, ..., end_effector link, gripper]
Found HDF5 file: ./benchmark1_1_release/h5_franka_fr3_dual/both_pour_water/success_episodes/train/1213_141924/data/trajectory.hdf5
File structure:
master (Group)
joint_position (Dataset)
Shape: (407, 16), Type: float64
observations (Group)
depth_images (Group)
camera_front (Dataset)
Shape: (407,), Type: object
camera_left (Dataset)
Shape: (407,), Type: object
camera_right (Dataset)
Shape: (407,), Type: object
camera_top (Dataset)
Shape: (407,), Type: object
rgb_images (Group)
camera_front (Dataset)
Shape: (407,), Type: object
camera_left (Dataset)
Shape: (407,), Type: object
camera_right (Dataset)
Shape: (407,), Type: object
camera_top (Dataset)
Shape: (407,), Type: object
puppet (Group)
end_effector (Dataset)
Shape: (407, 12), Type: float64
joint_position (Dataset)
Shape: (407, 16), Type: float64
Simulation Franka¶
Note¶
The h5 format is identical to that of the Simulation Franka described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
end_effector (7d): [x, y, z, quaternions (x,y,z,w)]
joint_position (8d): [base link, ..., end_effector link, gripper]
Note: In Isaac Sim simulation, we sample the robot state values at four times the camera frame rate. As a result, the timestep for the RGB images is about one-fourth that of the end-effector and joint position values.
Found HDF5 file: ./benchmark1_1_release/h5_sim_franka_3rgb/open_and_close_07/success_episodes/train/307342-2024_10_24_15_25_24/data/trajectory.hdf5
File structure:
franka (Group)
end_effector (Dataset)
Shape: (720, 7), Type: float64
joint_effort (Dataset)
Shape: (720, 8), Type: float64
joint_position (Dataset)
Shape: (720, 8), Type: float64
joint_velocity (Dataset)
Shape: (720, 8), Type: float64
observations (Group)
depth_images (Group)
camera_front_external (Dataset)
Shape: (176,), Type: object
camera_handeye (Dataset)
Shape: (176,), Type: object
camera_left_external (Dataset)
Shape: (176,), Type: object
camera_right_external (Dataset)
Shape: (176,), Type: object
rgb_images (Group)
camera_front_external (Dataset)
Shape: (170,), Type: object
camera_handeye (Dataset)
Shape: (170,), Type: object
camera_left_external (Dataset)
Shape: (170,), Type: object
camera_right_external (Dataset)
Shape: (170,), Type: object
sim_time_seq (Dataset)
Shape: (680,), Type: float64
Simulation Tien Kung¶
arm_joint_pos_seq (7d): [base link, ..., end_effector link]
hand_joint_pos_seq (12d): [thumb0 for bending, thumb1 for rotation, passive thumb0, passive thumb1, index finger, passive index finger, middle finger, passive middle finger, ring finger, passive ring finger, little finger, passive little finger]
For more detail of the inspire Dexterous hand, please refer to the dexterous hands instructions.
Found HDF5 file: ./benchmark1_1_release/h5_sim_tienkung_1rgb/pick_and_place_01_tienkung/success_episodes/train/1101314-2024_12_10_10_11_44/data/trajectory.hdf5
File structure:
observations (Group)
depth_images (Group)
camera_chest (Dataset)
Shape: (1363,), Type: object
camera_head (Dataset)
Shape: (1363,), Type: object
rgb_images (Group)
camera_chest (Dataset)
Shape: (1140,), Type: object
camera_head (Dataset)
Shape: (1140,), Type: object
sim_time_seq (Dataset)
Shape: (1140,), Type: float64
tiangong (Group)
left_arm_joint_effort_seqs (Dataset)
Shape: (1438, 7), Type: float64
left_arm_joint_pos_seq (Dataset)
Shape: (1436, 7), Type: float64
left_arm_joint_vel_seq (Dataset)
Shape: (1437, 7), Type: float64
left_end_effector_waist (Dataset)
Shape: (1438, 7), Type: float64
left_hand_joint_effort_seq (Dataset)
Shape: (1437, 12), Type: float64
left_hand_joint_pos_seq (Dataset)
Shape: (1436, 12), Type: float64
left_hand_joint_vel_seq (Dataset)
Shape: (1437, 12), Type: float64
right_arm_joint_effort_seq (Dataset)
Shape: (1438, 7), Type: float64
right_arm_joint_pos_seq (Dataset)
Shape: (1438, 7), Type: float64
right_arm_joint_vel_seq (Dataset)
Shape: (1438, 7), Type: float64
right_end_effector_waist (Dataset)
Shape: (1438, 7), Type: float64
right_hand_joint_pos_seq (Dataset)
Shape: (1438, 12), Type: float64
right_hand_joint_vel_seq (Dataset)
Shape: (1438, 12), Type: float64
Tien Kung Gello 1RGB¶
Note¶
The h5 format is identical to that of the Tien Kung Gello 1RGB described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
joint_position (16d): [left arm (7d), left hand closure (1d), right arm (7d), right hand closure (1d)]
arm (7d): [base link, ..., end_effector link]
Found HDF5 file: ./benchmark1_1_release/h5_tienkung_gello_1rgb/close_cabinet_drawer/success_episodes/train/1213_171743/data/trajectory.hdf5
File structure:
master (Group)
joint_position (Dataset)
Shape: (566, 16), Type: float64
observations (Group)
depth_images (Group)
camera_top (Dataset)
Shape: (566,), Type: object
rgb_images (Group)
camera_top (Dataset)
Shape: (566,), Type: object
puppet (Group)
joint_position (Dataset)
Shape: (566, 16), Type: float64
Tien Kung Prod1 Gello 1RGB¶
Note¶
The h5 format is identical to that of the Tien Kung Gello 1RGB described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
Found HDF5 file: ./benchmark1_1_release/h5_tienkung_prod1_gello_1rgb/clean_table/success_episodes/train/1216_144211/data/trajectory.hdf5
File structure:
master (Group)
joint_position (Dataset)
Shape: (660, 16), Type: float64
observations (Group)
depth_images (Group)
camera_top (Dataset)
Shape: (660,), Type: object
rgb_images (Group)
camera_top (Dataset)
Shape: (660,), Type: object
puppet (Group)
joint_position (Dataset)
Shape: (660, 16), Type: float64
Tien Kung Xsens 1RGB¶
Note¶
The h5 format is identical to that of the Tien Kung Xsens 1RGB described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
end_effector (12d): [left hand (6d), right hand (6d)]
hand (6d): [little finger, ring finger, middle finger, index finger, thumb0 for bending, thumb1 for rotation]
joint_position (14d): [left arm (7d), right arm (7d)]
arm (7d): [base link, ..., end_effector link]
Found HDF5 file: ./benchmark1_1_release/h5_tienkung_xsens_1rgb/pick_cup_pour_cup_place/success_episodes/train/2024-11-13-16-52-40/data/trajectory.hdf5
File structure:
master (Group)
end_effector (Dataset)
Shape: (662, 12), Type: float64
joint_position (Dataset)
Shape: (662, 14), Type: float64
observations (Group)
depth_images (Group)
camera_top (Dataset)
Shape: (662,), Type: object
rgb_images (Group)
camera_top (Dataset)
Shape: (662,), Type: object
puppet (Group)
end_effector (Dataset)
Shape: (662, 12), Type: float64
joint_position (Dataset)
Shape: (662, 14), Type: float64
UR 1RGB¶
Note¶
The h5 format is identical to that of the UR 1RGB described in
all_robot_h5_info.md（仓库内 v1.0 版参考）.
end_effector (6d): [x, y, z, r, p, y]
joint_position (7d): [base link, ..., end_effector link, gripper]
Found HDF5 file: ./benchmark1_1_release/h5_ur_1rgb/put_the_garbage_in_the_trash_can_1112/success_episodes/train/1112_171103/data/trajectory.hdf5
File structure:
master (Group)
joint_position (Dataset)
Shape: (203, 7), Type: float64
observations (Group)
depth_images (Group)
camera_top (Dataset)
Shape: (203,), Type: object
rgb_images (Group)
camera_top (Dataset)
Shape: (203,), Type: object
puppet (Group)
end_effector (Dataset)
Shape: (203, 6), Type: float64
joint_position (Dataset)
Shape: (203, 7), Type: float64