Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
name: string
scene_revision: int64
household_objects: bool
grasp_fit_depth: double
grasp_preload: double
motion_revision: int64
grasp_elevation_deg: int64
drawer_world_y: double
drawer_travel: double
top_down_grasp: bool
right_grasp_fit_depth: double
drawer_front_push: bool
drawer_push_style: string
drawer_parcel_x: double
drawer_parcel_y: double
right_grasp_preload: double
top_grasp_tilt_deg: int64
right_grasp_elevation_deg: int64
grasp_quality_max_slip_m: double
travel_blend_speed: double
approach_blend_speed: int64
handle_approach_blend_speed: int64
travel_feedback_tau: double
smooth_contact_transitions: bool
offset_range: double
target_offset_range: double
target_offset_xy: null
drawer_push_approach_clearance: double
drawer_push_approach_lift: double
handle_grasp_pitch_deg: int64
hand_position_range: list<item: double>
  child 0, item: double
hand_rotation_range_deg: list<item: double>
  child 0, item: double
hand_finger_range_rad: double
drawer_parcel_asset: string
drawer_release_z: double
bottle_upright_gain: double
right_top_grasp_style: string
right_grasp_back_offset: double
upright_objects: list<item: string>
  child 0, item: string
max_overlap_m: double
initial_hands_sha256: string
seconds: double
success: bool
failure: null
seed: int64
candidate: int64
task_progress: struct<opened: bool, released: bool, parcel_grasped: bool, placed: bool, closed: bool>
  child 0, opened: bool
  child 1, released: bool
  child 2, parcel_grasped: bool
  child 3, placed: bool
  child 4, closed: bool
task: string
grasp_quality_passed: bool
events: list<item: struct<time: double, phase: string, side: string, object: string>>
  child 0, item: struct<time: double, phase: string, side: string, object: string>
      child 0, time: double
      child 1, phase: string
      child 2, side: string
      child 3, object: string
grasp_quality: struct<parcel: struct<loaded_samples: int64, slip_max_m: double, rotation_drift_max_rad: double, opp (... 81 chars omitted)
  child 0, parcel: struct<loaded_samples: int64, slip_max_m: double, rotation_drift_max_rad: double, opposed_fraction:  (... 65 chars omitted)
      child 0, loaded_samples: int64
      child 1, slip_max_m: double
      child 2, rotation_drift_max_rad: double
      child 3, opposed_fraction: double
      child 4, thumb_load_p10_n: double
      child 5, other_digit_load_p10_n: double
hand_reset_sampling: struct<seed: int64, attempts: int64, distribution: string>
  child 0, seed: int64
  child 1, attempts: int64
  child 2, distribution: string
final_success_hold_controls: int64
to
{'seed': Value('int64'), 'task': Value('string'), 'success': Value('bool'), 'failure': Value('null'), 'final_success_hold_controls': Value('int64'), 'max_overlap_m': Value('float64'), 'initial_hands_sha256': Value('string'), 'hand_reset_sampling': {'seed': Value('int64'), 'attempts': Value('int64'), 'distribution': Value('string')}, 'seconds': Value('float64'), 'events': List({'time': Value('float64'), 'phase': Value('string'), 'side': Value('string'), 'object': Value('string')}), 'task_progress': {'opened': Value('bool'), 'released': Value('bool'), 'parcel_grasped': Value('bool'), 'placed': Value('bool'), 'closed': Value('bool')}, 'grasp_quality': {'parcel': {'loaded_samples': Value('int64'), 'slip_max_m': Value('float64'), 'rotation_drift_max_rad': Value('float64'), 'opposed_fraction': Value('float64'), 'thumb_load_p10_n': Value('float64'), 'other_digit_load_p10_n': Value('float64')}}, 'grasp_quality_passed': Value('bool'), 'candidate': Value('int64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 274, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3074, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2578, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2603, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 577, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 459, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 373, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2394, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2322, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              name: string
              scene_revision: int64
              household_objects: bool
              grasp_fit_depth: double
              grasp_preload: double
              motion_revision: int64
              grasp_elevation_deg: int64
              drawer_world_y: double
              drawer_travel: double
              top_down_grasp: bool
              right_grasp_fit_depth: double
              drawer_front_push: bool
              drawer_push_style: string
              drawer_parcel_x: double
              drawer_parcel_y: double
              right_grasp_preload: double
              top_grasp_tilt_deg: int64
              right_grasp_elevation_deg: int64
              grasp_quality_max_slip_m: double
              travel_blend_speed: double
              approach_blend_speed: int64
              handle_approach_blend_speed: int64
              travel_feedback_tau: double
              smooth_contact_transitions: bool
              offset_range: double
              target_offset_range: double
              target_offset_xy: null
              drawer_push_approach_clearance: double
              drawer_push_approach_lift: double
              handle_grasp_pitch_deg: int64
              hand_position_range: list<item: double>
                child 0, item: double
              hand_rotation_range_deg: list<item: double>
                child 0, item: double
              hand_finger_range_rad: double
              drawer_parcel_asset: string
              drawer_release_z: double
              bottle_upright_gain: double
              right_top_grasp_style: string
              right_grasp_back_offset: double
              upright_objects: list<item: string>
                child 0, item: string
              max_overlap_m: double
              initial_hands_sha256: string
              seconds: double
              success: bool
              failure: null
              seed: int64
              candidate: int64
              task_progress: struct<opened: bool, released: bool, parcel_grasped: bool, placed: bool, closed: bool>
                child 0, opened: bool
                child 1, released: bool
                child 2, parcel_grasped: bool
                child 3, placed: bool
                child 4, closed: bool
              task: string
              grasp_quality_passed: bool
              events: list<item: struct<time: double, phase: string, side: string, object: string>>
                child 0, item: struct<time: double, phase: string, side: string, object: string>
                    child 0, time: double
                    child 1, phase: string
                    child 2, side: string
                    child 3, object: string
              grasp_quality: struct<parcel: struct<loaded_samples: int64, slip_max_m: double, rotation_drift_max_rad: double, opp (... 81 chars omitted)
                child 0, parcel: struct<loaded_samples: int64, slip_max_m: double, rotation_drift_max_rad: double, opposed_fraction:  (... 65 chars omitted)
                    child 0, loaded_samples: int64
                    child 1, slip_max_m: double
                    child 2, rotation_drift_max_rad: double
                    child 3, opposed_fraction: double
                    child 4, thumb_load_p10_n: double
                    child 5, other_digit_load_p10_n: double
              hand_reset_sampling: struct<seed: int64, attempts: int64, distribution: string>
                child 0, seed: int64
                child 1, attempts: int64
                child 2, distribution: string
              final_success_hold_controls: int64
              to
              {'seed': Value('int64'), 'task': Value('string'), 'success': Value('bool'), 'failure': Value('null'), 'final_success_hold_controls': Value('int64'), 'max_overlap_m': Value('float64'), 'initial_hands_sha256': Value('string'), 'hand_reset_sampling': {'seed': Value('int64'), 'attempts': Value('int64'), 'distribution': Value('string')}, 'seconds': Value('float64'), 'events': List({'time': Value('float64'), 'phase': Value('string'), 'side': Value('string'), 'object': Value('string')}), 'task_progress': {'opened': Value('bool'), 'released': Value('bool'), 'parcel_grasped': Value('bool'), 'placed': Value('bool'), 'closed': Value('bool')}, 'grasp_quality': {'parcel': {'loaded_samples': Value('int64'), 'slip_max_m': Value('float64'), 'rotation_drift_max_rad': Value('float64'), 'opposed_fraction': Value('float64'), 'thumb_load_p10_n': Value('float64'), 'other_digit_load_p10_n': Value('float64')}}, 'grasp_quality_passed': Value('bool'), 'candidate': Value('int64')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Bimanual new tasks — preview episodes (v5)

One physically simulated human demonstration per task, shared for review before full data generation (400 demonstrations per task).

Common setup. Two floating articulated human hands (Superdex), MuJoCo physics at 600 Hz, scripted controller at 30 Hz, observations at 10 Hz. Randomization (between Round 10 and Round 11): object XY ±0.02 m, target (box / drawer) XY ±0.015 m, initial wrist XYZ ±(0.0175, 0.0175, 0.0125) m, wrist Euler ±(5, 5, 7.5)°, finger joints ±0.04 rad. Every episode is physically successful (success held ≥ 1 s, max penetration ≤ 6 mm) and passes the robot arm/wrist kinematic reachability check (limits: wrist position 35 mm, orientation 30°).

Task 1 — pick and place: bottle + Spam can (pick_place_bottle_spam/)

  • Left hand — bottle (natural side grasp): straight, natural wrist; the hand comes from the side, palm facing the bottle; thumb extended (not curled) on one side, index/middle/ring/little fingers around the other side. An upright-holding feedback keeps the bottle vertical; it is placed upright in the box, then the left hand backs out and returns to its start pose.
  • Right hand — Spam can (YCB 010_potted_meat_can): overhead hand orientation moved back, thumb on one side face, index/middle/ring/little together on the opposite side face; placed upright in the box.
  • Reachability reachability/00000.json: accepted, max wrist position error 18.8 mm, orientation 22.5°.
  • Controller success over 24 random seeds: 20/24 (the 4 failures are the Spam can slipping; failed attempts are discarded during generation).

Task 2 — drawer + Spam can (drawer_spam/)

  • Left hand grasps the original drawer handle with the original handle grasp and opens the drawer.
  • Right hand grasps the Spam can with the same grasp as Task 1, places it upright in the drawer, then pushes the drawer closed.
  • Reachability reachability/00001.json (same settings as the Round 11 drawer production: control-rate check, no posture reference): accepted, max wrist position error 33.8 mm, orientation 24.5°.
  • Controller success over 16 random seeds: 13/16 (the 3 failures are the Spam can slipping).

Task 2 update — drawer push hand yawed +25° (drawer_spam_push_yaw25/)

In production, the original closing push kept the robot's last wrist joint at its limit, so only ~3% of drawer attempts passed reachability. The right hand now pushes the drawer closed with its orientation rotated +25° about the vertical axis (everything else unchanged). Reachability acceptance among physically successful episodes rose to ~100% in sweeps. This episode is the first accepted one of the final drawer dataset (max wrist position error 14.4 mm, orientation 14.9°). The earlier drawer_spam/ preview is kept for reference.

Files (per task folder)

File Content
episode_XXXXX.h5 observations at 10 Hz: rgb (240×320), depth_m, depth_valid, pointcloud (2048×XYZRGB), point_valid, point_uv, proprio, actions (68D human hand commands: per hand xyz + quat wxyz + 27 finger joints), 30 Hz control_*, integration_state (full simulator state for exact replay); attrs: calibration, task_config, human_config, seed
episode_XXXXX_preview.mp4 recorded egocentric RGB, recorded depth, two external views re-rendered from the saved simulator states
episode_XXXXX.json, task_config.json, generation_task_config.json, scene.xml report and exact configuration
reachability/XXXXX.{json,npz} robot arm/wrist kinematic reachability evidence (not robot actions)
Downloads last month
22