Datasets:

jellyho
/

dexjoco-ogbench-30hz

Error code:   StreamingRowsError
Exception:    TypeError
Message:      Couldn't cast array of type string to null
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, 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 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, 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 483, 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 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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 343, 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 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2312, in cast_table_to_schema
                  cast_array_to_feature(
                  ~~~~~~~~~~~~~~~~~~~~~^
                      table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                      feature,
                      ^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
                  return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
                                           ~~~~^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2118, in cast_array_to_feature
                  casted_array_values = _c(array.values, feature.feature)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
                  return func(array, *args, **kwargs)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2152, in cast_array_to_feature
                  return array_cast(
                      array,
                  ...<2 lines>...
                      allow_decimal_to_str=allow_decimal_to_str,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
                  return func(array, *args, **kwargs)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2014, in array_cast
                  raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
              TypeError: Couldn't cast array of type string to null

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.

DexJoCo, in OGBench's format

The DexJoCo demonstrations replayed into the layout OGBench's loaders read: one row per STATE, terminals marking each trajectory's last row, and that row's action a pad.

Why a replay was necessary

The released recordings carry each object's pose once, at reset. Measured on water_plant demo 10, the fifteen *_ori_pose dimensions have a per-dimension range of exactly 0 across all 309 rows: they exist so a restorer can put a fresh scene back where a recording started, which is a different job from telling a policy where the object is now. So each demonstration is replayed and the live pose read at every step.

What an observation contains

proprioception | object poses (7 each) | object joints (1 each) | task progress (1 each), each block sorted by key, with the names and widths in the sidecar JSON so the privileged part can be sliced off again.

The last two blocks exist because an audit of all eleven _compute_success implementations found seven whose condition the observation could not reproduce. A hinge angle is a physical quantity that was simply not asked for; a passcode index is a counter the environment keeps in a Python attribute. Which counters are carried was measured rather than assumed: for each candidate, how long must a policy act while the variable sits at an intermediate value? unlock_index (3 changes, 58 steps at each) and pinch_count (3 changes, 126 steps) are carried; display_blue, screen_unlocked and trigger_pulled each turn on once and success follows after a debounce, so no policy ever acts while knowing them, and they are absent.

Rewards, and the failures

rewards is 0 until the step the environment reports success, then 1, and the episode ends there. episode_success labels each episode. Failed replays are kept: they are data, and a critic needs the contrast.

Actions

Stored unscaled -- metres, quaternion components and joint targets in one vector -- with action_low and action_high inside the npz, next to the numbers they apply to. Map to [-1, 1] with (a - low) / (high - low + 1e-8) * 2 - 1 and invert before stepping the environment.

Environment

These need github.com/jellyho/dexjoco, branch live-object-state: the live-pose hook, the joint and progress hooks, and the numpy-2 fixes without which no single environment can both train a policy and step this simulator.

task	episodes	success	replay	obs	proprio	object pose	joint	progress	action	rows
`bimanual_microwave_cook`	100	100	1.000	76	61	14	1	0	46	70540
`bimanual_photograph`	100	99	0.990	68	61	7	0	0	46	40355
`water_plant`	100	96	0.960	46	38	7	1	0	23	27463
`hammer_nail`	100	86	0.860	52	38	14	0	0	23	22039
`pinch_tongs`	100	82	0.820	39	31	7	0	1	23	43435
`click_mouse`	100	75	0.750	45	31	14	0	0	23	39845
`bimanual_assembly`	100	69	0.690	75	61	14	0	0	46	57926
`fold_glasses`	100	66	0.660	47	38	7	2	0	23	57610
`bimanual_hanoi`	100	63	0.630	78	50	28	0	0	46	109056
`bimanual_unlock_ipad`	100	57	0.570	69	61	7	0	1	46	44472

Downloads last month: 64

Total file size:

140 MB