Hugging Face's logo

Datasets:
THULab
/
sort_trash_real_2

Tasks:
Robotics
Modalities:
Time-series
Size:
10K<n<100K
Tags:
LeRobot
tsfile
timeseries
format:tsfile
Dataset card Files
xet
 Community
Episodes Preview LeRobot Visualizer
27 episodes · 50 fps

Sort Trash Real 2 (TsFile)

Converted from theconstruct-ai/sort_trash_real_2 at pinned revision ff96af085cd159ddb4d9e00bd83cb857bfe1fdd7. Modalities: Time-series.

Dataset description

This LeRobot v2.1 trajectory dataset records the source task sort the trash. Its numeric schema contains whole-body state/action signals plus motion-token, SMPL, planner, hand/wrist, and VR teleoperation features.

The pinned task table contains one instruction: “sort the trash”. The pinned metadata does not declare a robot type.

  • Source repository owner/publisher: theconstruct-ai
  • License: not declared by the pinned source; no license is asserted here.
  • Paper/homepage/citation: no completed paper, homepage, or citation is documented in the pinned source card unless linked above.

Source metadata note: The pinned source revision has no README.md; source facts come from its meta/info.json, task metadata, data Parquets, and config. No source license, paper, homepage, citation, or author list was declared.

Dataset Scale

Split Episodes Tasks Source trajectory Parquets TsFile rows Sampling rate TsFile files/shards
train 27 1 27 47,528 50 Hz 1

The staged Parquet has 47,528 rows and 388 columns including Time; TsFile chunk metadata independently reports the same 47,528 rows.

TsFile schema

Column Role TsFile type Observed/source range
Time TIME INT64 0–46,220 ms; restarts per episode
episode_index TAG STRING source integer 0–26
task_index TAG STRING source integer 0–0
frame_index FIELD INT64 0–2,311 within an episode
sample_index FIELD INT64 0–47,527 globally

Exact remaining FIELD names/ranges and imported types:

  • teleop_delta_heading (DOUBLE), teleop_smpl_frame_index (INT64), teleop_stream_mode, teleop_planner_mode (INT64), teleop_planner_speed, teleop_planner_height (FLOAT)
  • observation_state_0–_42, action_wbc_0–_42, observation_eef_state_0–_13 (FLOAT)
  • observation_root_orientation_0–_3, observation_projected_gravity_0–_2, observation_cpp_rotation_offset_0–_3, observation_init_base_quat_0–_3 (FLOAT)
  • action_motion_token_0–_63 (FLOAT)
  • teleop_smpl_joints_0–_71, teleop_smpl_pose_0–_62 (FLOAT)
  • teleop_body_quat_w_0–_3, teleop_target_body_orientation_0–_5 (FLOAT)
  • teleop_left_hand_joints_0–_6, teleop_right_hand_joints_0–_6, teleop_left_wrist_joints_0–_2, teleop_right_wrist_joints_0–_2 (FLOAT)
  • teleop_planner_movement_0–_2, teleop_planner_facing_0–_2 (FLOAT)
  • teleop_vr_3pt_position_0–_8, teleop_vr_3pt_orientation_0–_17 (FLOAT)

Conversion

  • All source episodes in the train split are merged into data/sort_trash_real_2_train.tsfile; episode_index and task_index are TAG dimensions.
  • Time = round(timestamp * 1000) in milliseconds. The source timestamp column is omitted because it is redundant with Time / 1000 seconds.
  • frame_index is retained. Source index is retained as sample_index.
  • Every vector is fully flattened: the complete source name is kept, . becomes _, and element indices are appended. Float vectors are imported as single-precision FLOAT fields.
  • Other scalar source columns shown above are retained; no trajectory rows are intentionally dropped.
  • Source metadata is mirrored for publication, with copied meta/info.json rewritten to describe the converted data path and conversion semantics.

Video policy

The pinned metadata declares 27 source videos for observation.images.ego_view. MP4 files were not downloaded or uploaded; they remain in the pinned source videos/ tree.

Minimal read example 

from tsfile import TsFileReader

reader = TsFileReader("data/sort_trash_real_2_train.tsfile")
print(reader.get_all_table_schemas().keys())
reader.close()

Source and provenance

Downloads last month
23
Total file size:
33.7 MB