Dataset Viewer
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
episode_id: string
outcome: string
history: list<item: struct<previous: string, outcome: string, success_score: double, time_ns: int64, labeler: (... 23 chars omitted)
child 0, item: struct<previous: string, outcome: string, success_score: double, time_ns: int64, labeler: string, no (... 11 chars omitted)
child 0, previous: string
child 1, outcome: string
child 2, success_score: double
child 3, time_ns: int64
child 4, labeler: string
child 5, note: string
success_score: double
metadata: struct<render: bool, method: string, profile: string, run_dir: string, step: string, files: struct</ (... 16108 chars omitted)
child 0, render: bool
child 1, method: string
child 2, profile: string
child 3, run_dir: string
child 4, step: string
child 5, files: struct</home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/flags.json: string, /home/junhyeong/FM_ (... 2104 chars omitted)
child 0, /home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/flags.json: string
child 1, /home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/params_0.pkl: string
child 2, /home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/norm_stats.json: string
child 3, /home/junhyeong/FM_RL/safety_profiles/bowl_h8_15hz.yaml: string
child 4, /home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/benchmark.py: string
child 5, /home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/te
...
child 11, variant: string
child 12, encoder: string
child 13, D_a: int64
child 14, D_c: int64
child 15, chunk: struct<H: int64, stride: int64, interp: int64, hold: int64>
child 0, H: int64
child 1, stride: int64
child 2, interp: int64
child 3, hold: int64
child 16, random_init: bool
child 17, replay: null
child 18, git: struct<fv: null, q: null>
child 0, fv: null
child 1, q: null
child 19, log_dir: string
child 20, warmup_refill_ms: double
child 21, model_hub: struct<arm: string, backend_port: int64, run_dir: string, profile: string, switches: int64>
child 0, arm: string
child 1, backend_port: int64
child 2, run_dir: string
child 3, profile: string
child 4, switches: int64
child 22, server_npz: string
stop_ns: int64
accepted_start: bool
status: string
error: null
clock: string
schema: string
start_wall_ns: int64
termination_reason: string
dropped: int64
start_ns: int64
counts: struct<lifecycle: int64, request: int64, cam1: int64, cam2: int64, reply: int64, leader_command: int (... 96 chars omitted)
child 0, lifecycle: int64
child 1, request: int64
child 2, cam1: int64
child 3, cam2: int64
child 4, reply: int64
child 5, leader_command: int64
child 6, gripper_command: int64
child 7, robot_state: int64
child 8, controller_command: int64
child 9, gripper_state: int64
display_timezone: string
to
{'schema': Value('string'), 'episode_id': Value('string'), 'status': Value('string'), 'clock': Value('string'), 'accepted_start': Value('bool'), 'start_ns': Value('int64'), 'stop_ns': Value('int64'), 'start_wall_ns': Value('int64'), 'display_timezone': Value('string'), 'termination_reason': Value('string'), 'dropped': Value('int64'), 'error': Value('null'), 'counts': {'lifecycle': Value('int64'), 'request': Value('int64'), 'cam1': Value('int64'), 'cam2': Value('int64'), 'reply': Value('int64'), 'leader_command': Value('int64'), 'gripper_command': Value('int64'), 'robot_state': Value('int64'), 'controller_command': Value('int64'), 'gripper_state': Value('int64')}, 'metadata': {'render': Value('bool'), 'method': Value('string'), 'profile': Value('string'), 'run_dir': Value('string'), 'step': Value('string'), 'files': {'/home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/flags.json': Value('string'), '/home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/params_0.pkl': Value('string'), '/home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/norm_stats.json': Value('string'), '/home/junhyeong/FM_RL/safety_profiles/bowl_h8_15hz.yaml': Value('string'), '/home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/benchmark.py': Value('string'), '/home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/test_svf_diagnostic.py': Value('string'), '/home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/prefl
...
g'), '/home/junhyeong/svf_real_carrot_commonenc_20260919/runtime/qflow_svf_merged/utils/encoders.py': Value('string')}, 'task': Value('string'), 'sampler': Value('string'), 'policy_network': Value('string'), 'serving_adapter': Value('string'), 'model_hub': {'arm': Value('string'), 'label': Value('string'), 'task': Value('string'), 'hub_pid': Value('int64'), 'hub_started_kst': Value('timestamp[s]'), 'backend_port': Value('int64'), 'router_port': Value('int64')}, 'recording_run_dir': Value('string'), 'accepted_start': Value('bool'), 'native_camera_min_spacing_ns': Value('int64'), 'reset_reply': {'ok': Value('bool'), 'policy_type': Value('string'), 'checkpoint': Value('string'), 'checkpoint_revision': Value('int64'), 'state_dim': Value('int64'), 'action_dim': Value('int64'), 'n_action_steps': Value('int64'), 'seed': Value('int64'), 'torch_seed': Value('int64'), 'reset_counter': Value('int64'), 'sampler': Value('string'), 'variant': Value('string'), 'encoder': Value('string'), 'D_a': Value('int64'), 'D_c': Value('int64'), 'chunk': {'H': Value('int64'), 'stride': Value('int64'), 'interp': Value('int64'), 'hold': Value('int64')}, 'random_init': Value('bool'), 'replay': Value('null'), 'git': {'fv': Value('null'), 'q': Value('null')}, 'log_dir': Value('string'), 'warmup_refill_ms': Value('float64'), 'model_hub': {'arm': Value('string'), 'backend_port': Value('int64'), 'run_dir': Value('string'), 'profile': Value('string'), 'switches': Value('int64')}}, 'server_npz': Value('string')}}
because column names don't match
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 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
episode_id: string
outcome: string
history: list<item: struct<previous: string, outcome: string, success_score: double, time_ns: int64, labeler: (... 23 chars omitted)
child 0, item: struct<previous: string, outcome: string, success_score: double, time_ns: int64, labeler: string, no (... 11 chars omitted)
child 0, previous: string
child 1, outcome: string
child 2, success_score: double
child 3, time_ns: int64
child 4, labeler: string
child 5, note: string
success_score: double
metadata: struct<render: bool, method: string, profile: string, run_dir: string, step: string, files: struct</ (... 16108 chars omitted)
child 0, render: bool
child 1, method: string
child 2, profile: string
child 3, run_dir: string
child 4, step: string
child 5, files: struct</home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/flags.json: string, /home/junhyeong/FM_ (... 2104 chars omitted)
child 0, /home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/flags.json: string
child 1, /home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/params_0.pkl: string
child 2, /home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/norm_stats.json: string
child 3, /home/junhyeong/FM_RL/safety_profiles/bowl_h8_15hz.yaml: string
child 4, /home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/benchmark.py: string
child 5, /home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/te
...
child 11, variant: string
child 12, encoder: string
child 13, D_a: int64
child 14, D_c: int64
child 15, chunk: struct<H: int64, stride: int64, interp: int64, hold: int64>
child 0, H: int64
child 1, stride: int64
child 2, interp: int64
child 3, hold: int64
child 16, random_init: bool
child 17, replay: null
child 18, git: struct<fv: null, q: null>
child 0, fv: null
child 1, q: null
child 19, log_dir: string
child 20, warmup_refill_ms: double
child 21, model_hub: struct<arm: string, backend_port: int64, run_dir: string, profile: string, switches: int64>
child 0, arm: string
child 1, backend_port: int64
child 2, run_dir: string
child 3, profile: string
child 4, switches: int64
child 22, server_npz: string
stop_ns: int64
accepted_start: bool
status: string
error: null
clock: string
schema: string
start_wall_ns: int64
termination_reason: string
dropped: int64
start_ns: int64
counts: struct<lifecycle: int64, request: int64, cam1: int64, cam2: int64, reply: int64, leader_command: int (... 96 chars omitted)
child 0, lifecycle: int64
child 1, request: int64
child 2, cam1: int64
child 3, cam2: int64
child 4, reply: int64
child 5, leader_command: int64
child 6, gripper_command: int64
child 7, robot_state: int64
child 8, controller_command: int64
child 9, gripper_state: int64
display_timezone: string
to
{'schema': Value('string'), 'episode_id': Value('string'), 'status': Value('string'), 'clock': Value('string'), 'accepted_start': Value('bool'), 'start_ns': Value('int64'), 'stop_ns': Value('int64'), 'start_wall_ns': Value('int64'), 'display_timezone': Value('string'), 'termination_reason': Value('string'), 'dropped': Value('int64'), 'error': Value('null'), 'counts': {'lifecycle': Value('int64'), 'request': Value('int64'), 'cam1': Value('int64'), 'cam2': Value('int64'), 'reply': Value('int64'), 'leader_command': Value('int64'), 'gripper_command': Value('int64'), 'robot_state': Value('int64'), 'controller_command': Value('int64'), 'gripper_state': Value('int64')}, 'metadata': {'render': Value('bool'), 'method': Value('string'), 'profile': Value('string'), 'run_dir': Value('string'), 'step': Value('string'), 'files': {'/home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/flags.json': Value('string'), '/home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/params_0.pkl': Value('string'), '/home/junhyeong/FM_RL/deploy/bowl_first50_h8_15hz_bc0/norm_stats.json': Value('string'), '/home/junhyeong/FM_RL/safety_profiles/bowl_h8_15hz.yaml': Value('string'), '/home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/benchmark.py': Value('string'), '/home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/test_svf_diagnostic.py': Value('string'), '/home/junhyeong/gello_software_jazzy/ros2_ur_ws/setup_jazzy/real_eval_recording_dev/prefl
...
g'), '/home/junhyeong/svf_real_carrot_commonenc_20260919/runtime/qflow_svf_merged/utils/encoders.py': Value('string')}, 'task': Value('string'), 'sampler': Value('string'), 'policy_network': Value('string'), 'serving_adapter': Value('string'), 'model_hub': {'arm': Value('string'), 'label': Value('string'), 'task': Value('string'), 'hub_pid': Value('int64'), 'hub_started_kst': Value('timestamp[s]'), 'backend_port': Value('int64'), 'router_port': Value('int64')}, 'recording_run_dir': Value('string'), 'accepted_start': Value('bool'), 'native_camera_min_spacing_ns': Value('int64'), 'reset_reply': {'ok': Value('bool'), 'policy_type': Value('string'), 'checkpoint': Value('string'), 'checkpoint_revision': Value('int64'), 'state_dim': Value('int64'), 'action_dim': Value('int64'), 'n_action_steps': Value('int64'), 'seed': Value('int64'), 'torch_seed': Value('int64'), 'reset_counter': Value('int64'), 'sampler': Value('string'), 'variant': Value('string'), 'encoder': Value('string'), 'D_a': Value('int64'), 'D_c': Value('int64'), 'chunk': {'H': Value('int64'), 'stride': Value('int64'), 'interp': Value('int64'), 'hold': Value('int64')}, 'random_init': Value('bool'), 'replay': Value('null'), 'git': {'fv': Value('null'), 'q': Value('null')}, 'log_dir': Value('string'), 'warmup_refill_ms': Value('float64'), 'model_hub': {'arm': Value('string'), 'backend_port': Value('int64'), 'run_dir': Value('string'), 'profile': Value('string'), 'switches': Value('int64')}}, 'server_npz': Value('string')}}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
SVF real-robot evaluation episodes
Every evaluation episode reported in the real-robot section of the SVF paper: the flow policies (Flow BC, IFQL, Q-Flow, SVF; task-specific, 15 Hz, 8-target action chunks, 20 trials per task and method) and the GR00T N1.7 pair (behavior-cloning fine-tuning vs. SVF).
Each episode folder is published whole: recording.h5 (policy requests/replies, joint states at 500 Hz, controller commands, gripper, leader commands, camera frames as JPEG payloads), cam1.mp4/cam2.mp4 (policy-input views), episode.json (recording metadata), outcome.json (operator label), placement_reference.json (object placement reference shown to the operator), summary.json, renders and logs.
Layout: <family>/<task>/<method>/<episode>/… with family in {flow, groot}. index.jsonl has one row per episode (task, method, outcome, start time, duration, placement reference, file list with sizes).
Success is strict (outcome == success); partial_success counts as failure in the paper.
| family | task | method | success / episodes | size |
|---|---|---|---|---|
| flow | Bowl stacking | Flow BC | 8 / 20 | 2.55 GB |
| flow | Bowl stacking | IFQL | 11 / 22 | 2.26 GB |
| flow | Bowl stacking | Q-Flow | 19 / 20 | 1.09 GB |
| flow | Bowl stacking | SVF | 19 / 20 | 0.98 GB |
| flow | Triple bowl stacking | Flow BC | 0 / 20 | 3.37 GB |
| flow | Triple bowl stacking | IFQL | 0 / 20 | 3.36 GB |
| flow | Triple bowl stacking | Q-Flow | 6 / 20 | 2.88 GB |
| flow | Triple bowl stacking | SVF | 6 / 20 | 3.00 GB |
| flow | Carrot in pot | Flow BC | 4 / 20 | 3.28 GB |
| flow | Carrot in pot | IFQL | 11 / 20 | 2.19 GB |
| flow | Carrot in pot | Q-Flow | 13 / 20 | 1.98 GB |
| flow | Carrot in pot | SVF | 16 / 19 | 1.48 GB |
| flow | Cube stacking | Flow BC | 0 / 20 | 3.61 GB |
| flow | Cube stacking | IFQL | 4 / 20 | 3.42 GB |
| flow | Cube stacking | Q-Flow | 5 / 20 | 3.08 GB |
| flow | Cube stacking | SVF | 10 / 20 | 5.50 GB |
| groot | Bowl stacking | GR00T N1.7 BC fine-tuning | 20 / 20 | 3.39 GB |
| groot | Bowl stacking | GR00T N1.7 + SVF | 20 / 20 | 3.23 GB |
| groot | Triple bowl stacking | GR00T N1.7 BC fine-tuning | 18 / 20 | 9.21 GB |
| groot | Triple bowl stacking | GR00T N1.7 + SVF | 20 / 20 | 9.44 GB |
| groot | Carrot in pot | GR00T N1.7 BC fine-tuning | 19 / 20 | 7.05 GB |
| groot | Carrot in pot | GR00T N1.7 + SVF | 19 / 20 | 5.75 GB |
| groot | Cube stacking | GR00T N1.7 BC fine-tuning | 1 / 20 | 10.33 GB |
| groot | Cube stacking | GR00T N1.7 + SVF | 20 / 20 | 4.03 GB |
Total: 481 episodes, 96.4 GB. Generated 2026-09-26 00:07 KST.
- Downloads last month
- 157