RoboSynChallenge/cobotmagic_Real_click_bell
CobotMagic Real Dataset: click_bell Generated by convert_piper_hdf5_to_lerobot.py. RoboSynChallenge Alignment Task dir: click_bell Environment ID: ClickBell Real dataset: RoboSynChallenge/cobotmagic_Real_click_bell Aligned simulation dataset: click_bell/cobotmagic_Sim_click_the_bell_000 Prompt: Click the bell Summary Robot type: aloha FPS: 10 Sampling mode: full Leading group: slave_cam_high Crop strategy: LATEST_START Default leading tolerance:… See the full description on the dataset page: https://huggingface.co/datasets/RoboSynChallenge/cobotmagic_Real_click_bell.
CobotMagic Real Dataset: click_bell
Generated by convert_piper_hdf5_to_lerobot.py.
RoboSynChallenge Alignment
- Task dir:
click_bell - Environment ID:
ClickBell - Real dataset:
RoboSynChallenge/cobotmagic_Real_click_bell - Aligned simulation dataset:
click_bell/cobotmagic_Sim_click_the_bell_000 - Prompt:
Click the bell
Summary
- Robot type:
aloha - FPS:
10 - Sampling mode:
full - Leading group:
slave_cam_high - Crop strategy:
LATEST_START - Default leading tolerance:
0.1s - 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.
Notes
- Timestamp alignment is done per HDF5 group. Each camera group and each arm group has its own
timestampdataset, which is treated like an individual topic timeline. - The leading group is
slave_cam_high. Infullmode every leading-group timestep is kept; infpsmode 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. The historical controller saved gripper as
mm / 70; this dataset stores gripper asmm / 100, i.e. a normalized physical opening ratio where 0 is closed and 1 is about 100 mm open. ee_poserotation is decoded from the current PiperController convention: controller value -> multiply by 1000 -> Piper xyz Euler degrees -> convert torot6d. WhenEE_ROT_FORMATisrpy/euler, the exported values are radians.GRIPPER_FORMAT="01"treats values in[0, 1]as normalized opening ratio and thresholds them withGRIPPER_FULL_WIDTH=0.05andGRIPPER_THRESHOLD=0.01.
Image Keys
Feature Names
observation.state / action names:
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