Datasets:

origin-data-lab
/

egocentric-rgb-imu-sample

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema: string
timebase: string
annotation_method: string
task: string
segments: list<item: struct<start_s: double, end_s: double, label: string, objects: list<item: string>, intera (... 15 chars omitted)
  child 0, item: struct<start_s: double, end_s: double, label: string, objects: list<item: string>, interaction: stri (... 3 chars omitted)
      child 0, start_s: double
      child 1, end_s: double
      child 2, label: string
      child 3, objects: list<item: string>
          child 0, item: string
      child 4, interaction: string
boundary_note: string
events: list<item: struct<start_s: double, end_s: double, actor: string, action: string, tool: string, targe (... 11 chars omitted)
  child 0, item: struct<start_s: double, end_s: double, actor: string, action: string, tool: string, target: string>
      child 0, start_s: double
      child 1, end_s: double
      child 2, actor: string
      child 3, action: string
      child 4, tool: string
      child 5, target: string
note: string
spatial_annotation_included: bool
field_definitions: struct<start_s: string, end_s: string, actor: string, action: string, tool: string, target: string>
  child 0, start_s: string
  child 1, end_s: string
  child 2, actor: string
  child 3, action: string
  child 4, tool: string
  child 5, target: string
to
{'schema': Value('string'), 'timebase': Value('string'), 'spatial_annotation_included': Value('bool'), 'field_definitions': {'start_s': Value('string'), 'end_s': Value('string'), 'actor': Value('string'), 'action': Value('string'), 'tool': Value('string'), 'target': Value('string')}, 'events': List({'start_s': Value('float64'), 'end_s': Value('float64'), 'actor': Value('string'), 'action': Value('string'), 'tool': Value('string'), 'target': Value('string')}), 'note': 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 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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
              schema: string
              timebase: string
              annotation_method: string
              task: string
              segments: list<item: struct<start_s: double, end_s: double, label: string, objects: list<item: string>, intera (... 15 chars omitted)
                child 0, item: struct<start_s: double, end_s: double, label: string, objects: list<item: string>, interaction: stri (... 3 chars omitted)
                    child 0, start_s: double
                    child 1, end_s: double
                    child 2, label: string
                    child 3, objects: list<item: string>
                        child 0, item: string
                    child 4, interaction: string
              boundary_note: string
              events: list<item: struct<start_s: double, end_s: double, actor: string, action: string, tool: string, targe (... 11 chars omitted)
                child 0, item: struct<start_s: double, end_s: double, actor: string, action: string, tool: string, target: string>
                    child 0, start_s: double
                    child 1, end_s: double
                    child 2, actor: string
                    child 3, action: string
                    child 4, tool: string
                    child 5, target: string
              note: string
              spatial_annotation_included: bool
              field_definitions: struct<start_s: string, end_s: string, actor: string, action: string, tool: string, target: string>
                child 0, start_s: string
                child 1, end_s: string
                child 2, actor: string
                child 3, action: string
                child 4, tool: string
                child 5, target: string
              to
              {'schema': Value('string'), 'timebase': Value('string'), 'spatial_annotation_included': Value('bool'), 'field_definitions': {'start_s': Value('string'), 'end_s': Value('string'), 'actor': Value('string'), 'action': Value('string'), 'tool': Value('string'), 'target': Value('string')}, 'events': List({'start_s': Value('float64'), 'end_s': Value('float64'), 'actor': Value('string'), 'action': Value('string'), 'tool': Value('string'), 'target': Value('string')}), 'note': Value('string')}
              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.

Origin Data Lab — Egocentric RGB + IMU Evaluation Sample

Version 1.0.3 · First-party technical evaluation sample · Technical Field Note #02

This dataset is a compact public capability demonstration from Origin Data Lab for evaluating an egocentric RGB + IMU capture and delivery workflow.

It contains a short real-world task excerpt together with native and application-level IMU streams, structured task annotations, sanitized capture metadata, integrity records, checksums, and a reproducible Python verifier.

This sample demonstrates a specific released capture package. It does not imply that every Origin Data Lab production project uses the same hardware, rates, schema, or timing method.

Quick Facts

Item	Released sample
Video	1920 × 1080
Frame rate	30 fps
Frames	587
Duration	19.567 s
Native accelerometer	8,028 samples · 410.641 Hz
Native gyroscope	8,028 samples · 410.641 Hz
App-level accelerometer	2,001 samples · 102.353 Hz
App-level gyroscope	1,998 samples · 102.162 Hz
Audio	Not included
Task category	Food preparation
Environment	Domestic kitchen
Capture country	South Korea

The video source was portrait (1080 × 1920) and the public preview was transformed 90° counterclockwise without crop or stretch. IMU axes were not rotated to match the transformed preview.

Repository Contents

capture/
  video_preview_aligned_19p55s.mp4

sensors/
  app_level_accelerometer_clip.csv
  app_level_gyroscope_clip.csv
  native_accelerometer_clip.csv
  native_gyroscope_clip.csv

annotations/
  interaction_events.json
  task_structure.json

metadata/
  capture_metadata_public.json

quality/
  integrity_report.json
  verification_output.txt

tools/
  verify_odl_sample.py

checksums.sha256
DATA_CARD.md
LICENSE.md
README.md

Sensor Measurements

The released verifier computes stream statistics from sensor_timestamp_ns.

Stream	Samples	Measured rate	Max interval
App-level accelerometer	2,001	102.353 Hz	21.9177 ms
App-level gyroscope	1,998	102.162 Hz	21.9177 ms
Native accelerometer	8,028	410.641 Hz	2.4360 ms
Native gyroscope	8,028	410.641 Hz	2.4360 ms

Cross-stream consistency

For timestamps shared by the application-level and native streams:

accelerometer timestamp matches: 2,001 / 2,001
native/app accelerometer scale ratio: 9.806650–9.806651
gyroscope timestamp matches: 1,998 / 1,998
native/app gyroscope value ratio: 1.000000

These are measured values from this released excerpt, not nominal device specifications.

Temporal Alignment

RGB and IMU in this public excerpt are software-aligned.

The source session used a software visual–inertial refinement of −446 ms before the public timeline was released.

Visual–inertial checks after alignment produced:

Check	Result
Visual–inertial Method A	0 ms · R² 0.798
Visual–inertial Method B	+6 ms · Pearson r 0.929
Method A quarter-window shifts	−4, −2, 0, +2 ms
Stated alignment uncertainty	within one 30 fps video frame (33 ms)

These checks are in-sample visual–inertial checks on the released excerpt.

This sample does not claim:

hardware synchronization between RGB and IMU;
guaranteed frame-level synchronization;
sub-frame synchronization;
held-out validation of the alignment method.

Coordinate Frame and Units

Public metadata identifies the IMU coordinate frame as:

Android device default; portrait source reference

The IMU axes were not rotated to match the landscape public preview.

For this release:

application-level accelerometer values are represented in g;
native accelerometer values are strongly supported as m/s² by the exact 9.80665 cross-stream scale relationship;
gyroscope values are represented as rad/s;
native and application-level gyroscope values are identical at matching timestamps.

See metadata/capture_metadata_public.json and DATA_CARD.md for the release-specific caveats.

Structured Task Metadata

The package includes:

annotations/task_structure.json
annotations/interaction_events.json

These files provide structured task and interaction information for the released excerpt. They are evaluation examples, not universal ontology definitions.

Integrity and Reproducibility

checksums.sha256 provides SHA-256 hashes for the released payload.

The included verifier checks:

SHA-256 manifest integrity
per-stream timestamp statistics
duplicate and non-monotonic timestamps
application/native cross-stream consistency
visual–inertial lag between image motion and native gyroscope motion

Run the verifier

pip install numpy opencv-python
python tools/verify_odl_sample.py .

A recorded verifier run is included at:

quality/verification_output.txt

The companion GitHub repository provides the verification tooling and technical documentation:

GitHub: https://github.com/origin-data-lab/egocentric-rgb-imu-sample

Privacy

The public package removes or excludes:

audio
exact location
collector identifiers
absolute Unix arrival timestamps
internal capture identifiers

Public field-data releases remain subject to applicable rights, consent, privacy, and project-specific requirements.

Intended Use

This sample is intended for:

technical evaluation of the released RGB + IMU package;
inspection of measured sensor characteristics;
inspection of timing methodology and limitations;
evaluation of task / interaction metadata structure;
integrity and reproducibility checks;
buyer-side discussion of custom Physical AI and robotics data requirements.

It should not be interpreted as a benchmark dataset or as proof of suitability for a particular robot policy, VLA system, world model, or downstream training pipeline without buyer-specific evaluation.

Cite

@misc{origindatalab2026fieldnote02,
  author       = {{Origin Data Lab}},
  title        = {Egocentric RGB + IMU Task Sample},
  howpublished = {Technical Field Note \#02},
  year         = {2026},
  url          = {https://origindatalab.io/technical-field-notes/egocentric-rgb-imu-task-sample.html}
}

License

This dataset uses the release-specific terms in LICENSE.md.

The license is intentionally marked as other in the Hugging Face metadata because the package uses its own evaluation terms rather than a standard open-source dataset license.

Origin Data Lab

Origin Data Lab produces real-world multimodal training data for Physical AI, robotics, and embodied AI, including field sourcing, worker onboarding, capture operations, sensor and metadata pipelines, QC, recapture, and customer-specific structured delivery.

Technical Field Note:
https://origindatalab.io/technical-field-notes/egocentric-rgb-imu-task-sample.html

All Technical Field Notes:
https://origindatalab.io/technical-field-notes/

Company:
https://origindatalab.io/

Release: v1.0.3
Sample type: capability demonstration
Validation scope: in-sample visual–inertial checks

Downloads last month: 45

Total file size:

30.5 MB