WorldCode Exported JSON Field Specification
The following specification is based on the actual outputs produced by the export script, as well as the example file worldcode_HS-2-1420125020208_2025-10-21-14-50-09_3_s0_vlta_reorg_sample_1-2.json.
Top-level Fields
| Field |
Description |
worldcode_name |
Name of the current sample, typically also the filename of the exported JSON. |
dataset_path |
Absolute path to the original worldcode directory. |
task_description |
Description of the task. |
status |
Overall status of the current worldcode, e.g., success. |
subtasks |
List of subtasks describing key action intervals. |
num_frames |
Total number of frames. |
camera_calibration |
Calibration parameters for each camera. |
frames |
Per-frame data. |
_validation_warnings |
Optional field; appears only when missing files are detected during validation. |
`subtasks`
Each subtask contains:
| Field |
Description |
subtask_index |
Subtask number. |
subtask_description |
Description of the subtask. |
status |
Status of the subtask. |
timestamp_range |
Timestamp range corresponding to the subtask. |
frame_index_range |
Frame index range corresponding to the subtask. |
`camera_calibration`
Organized by camera name, for example lf_chest_fisheye, ldr_hand_fisheye.
Each camera contains:
| Field |
Description |
intrinsic |
3x3 camera intrinsic matrix. |
distortion |
Distortion parameters. |
width |
Image width. |
height |
Image height. |
camera_model |
Camera model, e.g., fisheye. |
`frames`
Each frame contains:
| Field |
Description |
frame_index |
Frame index. |
timestamp |
Timestamp of the current frame. |
images |
Relative paths to images from each camera. |
wrist_poses |
Placeholder zero poses for the left and right end-effectors. |
chest_poses |
Poses of the left and right end-effectors in the current frame's chest coordinate system. |
ego_poses |
Pose of each camera relative to its own frame 0. |
chest_0_poses |
Pose of each camera in the chest coordinate system. |
hand_states |
Hand pose data; only populated when end_effector_type == "hand". |
Pose Fields
`wrist_poses`
- Meaning: Zero poses of
left_eef and right_eef.
- Values:
position: [0, 0, 0], quaternion: [0, 0, 0, 1].
- Physical meaning: Placeholder field representing the zero pose of the robotic arm end-effector (wrist).
`chest_poses`
- Meaning: Pose of
left_eef and right_eef relative to the chest frame at the current timestep.
- Computation:
chest_inv @ eef_pose_in_chest.
- Reference frame: Current frame's
chest.
- Physical meaning: Pose of the robotic arm end-effector in the current local
chest coordinate system.
`ego_poses`
- Meaning: Pose of each camera relative to its own frame 0.
- Computation:
base_inv @ curr_pose.
- Reference frame: Each camera's own frame 0.
- Physical meaning: Self-motion trajectory of the camera.
`chest_0_poses`
- Meaning: Pose of each camera in the
chest frame at frame 0.
- Computation: Read directly from
pose_in_chest in the HDF5 file.
- Reference frame:
chest.
- Physical meaning: Transform of the camera relative to the
chest.
Additional Field Notes
`images`
- Keys are camera names, values are relative paths to images.
- Paths are relative to the current worldcode root directory, e.g.,
camera/lf_chest_fisheye/xxx.jpg.
`hand_states`
- Used when
end_effector_type == "hand".
- Contains
left_hand and right_hand.
- Each value is the hand state vector for the current frame, taken directly from the data path specified in the configuration.