Datasets:

HumanoidEverywhere
/

amo-trajectories

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
T: int64
replay_steps: int64
score: struct<thermos: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>, vitamin (... 239 chars omitted)
  child 0, thermos: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>
      child 0, pos: list<item: double>
          child 0, item: double
      child 1, in_xy: bool
      child 2, below_rim: bool
      child 3, in_bin: bool
  child 1, vitamin_box: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>
      child 0, pos: list<item: double>
          child 0, item: double
      child 1, in_xy: bool
      child 2, below_rim: bool
      child 3, in_bin: bool
  child 2, clock: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>
      child 0, pos: list<item: double>
          child 0, item: double
      child 1, in_xy: bool
      child 2, below_rim: bool
      child 3, in_bin: bool
  child 3, pelvis_z: double
  child 4, min_pelvis_z: double
  child 5, upright: bool
  child 6, never_fell: bool
objects_total: int64
bin_aabb_lo: list<item: double>
  child 0, item: double
inside_max_z: double
robot_pelvis_z: double
bin_aabb_hi: list<item: double>
  child 0, item: double
never_fell: bool
fall_min_z: double
robot_upright: bool
per_object: struct<103: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, b (... 278 chars omitted)
  child 0, 103: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, below_rim: bo (... 17 chars omitted)
      child 0, label: string
      child 1, final_pos: list<item: double>
          child 0, item: double
      child 2, start_z: double
      child 3, in_bin_xy: bool
      child 4, below_rim: bool
      child 5, in_bin: bool
  child 1, 105: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, below_rim: bo (... 17 chars omitted)
      child 0, label: string
      child 1, final_pos: list<item: double>
          child 0, item: double
      child 2, start_z: double
      child 3, in_bin_xy: bool
      child 4, below_rim: bool
      child 5, in_bin: bool
  child 2, 106: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, below_rim: bo (... 17 chars omitted)
      child 0, label: string
      child 1, final_pos: list<item: double>
          child 0, item: double
      child 2, start_z: double
      child 3, in_bin_xy: bool
      child 4, below_rim: bool
      child 5, in_bin: bool
objects_in_bin: int64
min_pelvis_z: double
upright_min_z: double
success: bool
to
{'success': Value('bool'), 'score': Value('float64'), 'objects_in_bin': Value('int64'), 'objects_total': Value('int64'), 'robot_upright': Value('bool'), 'robot_pelvis_z': Value('float64'), 'never_fell': Value('bool'), 'min_pelvis_z': Value('float64'), 'per_object': {'103': {'label': Value('string'), 'final_pos': List(Value('float64')), 'start_z': Value('float64'), 'in_bin_xy': Value('bool'), 'below_rim': Value('bool'), 'in_bin': Value('bool')}, '105': {'label': Value('string'), 'final_pos': List(Value('float64')), 'start_z': Value('float64'), 'in_bin_xy': Value('bool'), 'below_rim': Value('bool'), 'in_bin': Value('bool')}, '106': {'label': Value('string'), 'final_pos': List(Value('float64')), 'start_z': Value('float64'), 'in_bin_xy': Value('bool'), 'below_rim': Value('bool'), 'in_bin': Value('bool')}}, 'bin_aabb_lo': List(Value('float64')), 'bin_aabb_hi': List(Value('float64')), 'inside_max_z': Value('float64'), 'upright_min_z': Value('float64'), 'fall_min_z': Value('float64'), 'replay_steps': 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 272, 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 489, 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 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              T: int64
              replay_steps: int64
              score: struct<thermos: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>, vitamin (... 239 chars omitted)
                child 0, thermos: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>
                    child 0, pos: list<item: double>
                        child 0, item: double
                    child 1, in_xy: bool
                    child 2, below_rim: bool
                    child 3, in_bin: bool
                child 1, vitamin_box: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>
                    child 0, pos: list<item: double>
                        child 0, item: double
                    child 1, in_xy: bool
                    child 2, below_rim: bool
                    child 3, in_bin: bool
                child 2, clock: struct<pos: list<item: double>, in_xy: bool, below_rim: bool, in_bin: bool>
                    child 0, pos: list<item: double>
                        child 0, item: double
                    child 1, in_xy: bool
                    child 2, below_rim: bool
                    child 3, in_bin: bool
                child 3, pelvis_z: double
                child 4, min_pelvis_z: double
                child 5, upright: bool
                child 6, never_fell: bool
              objects_total: int64
              bin_aabb_lo: list<item: double>
                child 0, item: double
              inside_max_z: double
              robot_pelvis_z: double
              bin_aabb_hi: list<item: double>
                child 0, item: double
              never_fell: bool
              fall_min_z: double
              robot_upright: bool
              per_object: struct<103: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, b (... 278 chars omitted)
                child 0, 103: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, below_rim: bo (... 17 chars omitted)
                    child 0, label: string
                    child 1, final_pos: list<item: double>
                        child 0, item: double
                    child 2, start_z: double
                    child 3, in_bin_xy: bool
                    child 4, below_rim: bool
                    child 5, in_bin: bool
                child 1, 105: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, below_rim: bo (... 17 chars omitted)
                    child 0, label: string
                    child 1, final_pos: list<item: double>
                        child 0, item: double
                    child 2, start_z: double
                    child 3, in_bin_xy: bool
                    child 4, below_rim: bool
                    child 5, in_bin: bool
                child 2, 106: struct<label: string, final_pos: list<item: double>, start_z: double, in_bin_xy: bool, below_rim: bo (... 17 chars omitted)
                    child 0, label: string
                    child 1, final_pos: list<item: double>
                        child 0, item: double
                    child 2, start_z: double
                    child 3, in_bin_xy: bool
                    child 4, below_rim: bool
                    child 5, in_bin: bool
              objects_in_bin: int64
              min_pelvis_z: double
              upright_min_z: double
              success: bool
              to
              {'success': Value('bool'), 'score': Value('float64'), 'objects_in_bin': Value('int64'), 'objects_total': Value('int64'), 'robot_upright': Value('bool'), 'robot_pelvis_z': Value('float64'), 'never_fell': Value('bool'), 'min_pelvis_z': Value('float64'), 'per_object': {'103': {'label': Value('string'), 'final_pos': List(Value('float64')), 'start_z': Value('float64'), 'in_bin_xy': Value('bool'), 'below_rim': Value('bool'), 'in_bin': Value('bool')}, '105': {'label': Value('string'), 'final_pos': List(Value('float64')), 'start_z': Value('float64'), 'in_bin_xy': Value('bool'), 'below_rim': Value('bool'), 'in_bin': Value('bool')}, '106': {'label': Value('string'), 'final_pos': List(Value('float64')), 'start_z': Value('float64'), 'in_bin_xy': Value('bool'), 'below_rim': Value('bool'), 'in_bin': Value('bool')}}, 'bin_aabb_lo': List(Value('float64')), 'bin_aabb_hi': List(Value('float64')), 'inside_max_z': Value('float64'), 'upright_min_z': Value('float64'), 'fall_min_z': Value('float64'), 'replay_steps': 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.

AMO whole-body pick-and-place trajectories (Unitree G1 + Dex3)

Generated whole-body trajectories for a Unitree G1 humanoid with Dex3 hands, driven by the AMO RL balance policy in MuJoCo. The robot walks to a table, grasps an object, carries it to a floor bin and drops it in.

Produced with the pipeline in Junyingw/simple-3dgs → scripts/amo_trajgen/ (see the Agent Trajectory Generation section of that repo's README for how to regenerate or adapt these).

vitamin_box_to_bin/          verified solution: vitamin box -> garbage bin
  trajectory.npz             the trajectory (see schema below)
  qpos_log.npz               per-frame qpos + step index (stride 5), for re-rendering
  summary.json               phase-level result summary
  grader_result.json         official grader verdict
  pickplace.mp4              38.8 s, box-tracking + chase cam, bin AABB overlaid

reference_replay/            the task's own reference trajectory, replayed
  qpos_log.npz
  grader_result.json
  replay_in_room.mp4         2:11, up to where the robot leaves the room

`trajectory.npz` schema

Four float arrays sharing the same first dimension T:

array	shape	meaning
`action`	`[T,43]`	per-joint targets in `robot_params.json["joints_names"]` order; `0:15` legs+waist, `15:29` arms, `29:43` the two Dex3 hands
`amo_policy_command`	`[T,9]`	`[0]` forward vel, `[1]` lateral vel, `[3]` torso yaw, `[4]` torso pitch, `[5]` torso roll, `[8]` body height (0.75 nominal)
`amo_policy_target_yaw`	`[T,1]`	target heading relative to the robot's initial facing, radians
`amo_policy_turning_flag`	`[T,1]`	1.0 while turning in place

The last row is never executed — it only terminates the sequence.

Verified result — `vitamin_box_to_bin`

vitamin_box final  [1.4395, 0.3516, 0.0805]      in_bin = True
robot upright      True    min_pelvis_z = 0.6562 (fall limit 0.45)
never fell         True    5811 replay steps

Reproduces identically across repeated independent open-loop grader runs. Margins are deliberate, not marginal: ≥8 cm inside the bin's XY bounds, 20 cm below the rim limit, 0.21 m above the fall threshold.

About `reference_replay`

The reference trajectory shipped with the task does not solve it as replayed here: it scores 0.0. thermos and clock are never touched (max displacement 1 mm), vitamin_box is knocked to the floor at step ~17,100, and from step ~19,620 the robot walks out of the room entirely, ending 29 m away — while never falling. Included as a baseline and because it contains one successful grasp, which the pipeline mines as a reusable primitive.

The replay contract is open-loop on velocity/heading commands, so heading error integrates without bound. That is why trajectories must be generated closed-loop against the simulator and only then verified by an open-loop replay.

Usage

import numpy as np
from huggingface_hub import hf_hub_download

p = hf_hub_download("HumanoidEverywhere/amo-trajectories",
                    "vitamin_box_to_bin/trajectory.npz", repo_type="dataset")
z = np.load(p)
print({k: z[k].shape for k in z.files})
# {'action': (5812, 43), 'amo_policy_command': (5812, 9),
#  'amo_policy_target_yaw': (5812, 1), 'amo_policy_turning_flag': (5812, 1)}

Replaying these requires the task assets (scene.mjb, init_state.npy, amo_init_state.npz, robot_params.json, and the AMO policy weights), which are not in this dataset. scene.mjb only loads in mujoco 3.3.x.

Downloads last month: 24

Total file size:

58.7 MB

AMO whole-body pick-and-place trajectories (Unitree G1 + Dex3)

Contents

trajectory.npz schema

Verified result — vitamin_box_to_bin

About reference_replay

Usage

`trajectory.npz` schema

Verified result — `vitamin_box_to_bin`

About `reference_replay`