π¦ Data Format v1 (Deprecated)#
This document describes the legacy v1 trajectory format previously used in RoboVerse and related frameworks. This format is now deprecated in favor of the more extensible Data Format v2.
π Typical Filename#
trajectory-unified.pkl
π Data Structure#
{
'source': 'ManiSkill2-rigid_body-PickSingleYCB-v0',
'max_episode_len': 147,
'demos': {
"franka": [
{
'name': 'banana_traj_0',
'description': 'Banana nana bana ba banana.', # optional
'env_setup': {
"init_q": [...], # [q_len], e.g., 9 for Franka
"init_robot_pos": [...], # [3]
"init_robot_quat": [...], # [4]
"init_{obj_name}_pos": [...], # [3]
"init_{obj_name}_quat": [...], # [4]
"init_{joint_name}_q": [...], # [1], for articulations
...
},
'robot_traj': {
"q": [[...], ...], # [demo_len x q_len]
"ee_act": [[...], ...], # [demo_len x 1] (0 = close, 1 = open)
}
},
...
],
}
}
π Field Descriptions#
Key |
Description |
---|---|
|
Environment string identifier |
|
Used to pad all trajectories for batching |
|
Dictionary mapping robot names to a list of demo dicts |
|
Initial states of robot and objects in the environment |
|
Joint positions for each timestep |
|
End-effector action per timestep (open/close gripper) |
In each env_setup
:
init_q
: Initial robot joint configurationinit_robot_pos
: Base robot positioninit_robot_quat
: Base robot orientation (quaternion)init_*
: Initial position/quaternion/state of relevant scene objects
These values are consumed inside the taskβs _reset_idx
function.
β οΈ Status#
This format is no longer supported in the latest RoboVerse infrastructure. If you have data in this format, use the provided script to convert it to v2:
python scripts/convert_traj_v1_to_v2.py --task CloseBox --robot franka