ViaCatalyst/robomimic-lift-ph-lerobot-v3

robomimic Lift PH — verified LeRobot v3 reference subset

Community conversion produced by ViaCatalyst BYOD. This repository is not an official upstream release and is not affiliated with the robomimic authors.

This is a compact, provenance-complete conversion of the first 10 episodes from the pinned robomimic Lift PH low-dimensional v1.5 HDF5 file. It is intended as a reproducible LeRobot v3 reference dataset and conversion-quality example.

At a glance

Property	Value
LeRobot format	v3.0
Robot	Franka Emika Panda
Task	Lift the object
Episodes	10
Frames	531
Duration	26.55 seconds
FPS	20
Observation modality	Low-dimensional state
Action dimension	7
State dimension	53
Official LeRobot reader	Passed with LeRobot 0.6.0

Features and source mapping

LeRobot feature	dtype	shape	Source mapping
action	float32	[7]	Direct per-frame copy from each demonstration's actions dataset, with a float32 cast only
observation.state	float32	[53]	Deterministic concatenation of numeric obs/* vectors documented by the feature names in meta/info.json
episode_index	int64	[1]	Preserved HDF5 demonstration boundary
frame_index	int64	[1]	Zero-based frame position inside each episode
timestamp	float32	[1]	frame_index / 20 seconds
task_index	int64	[1]	Maps to Lift the object in meta/tasks.parquet

Original-action preservation

The conversion reads the original actions array for each selected demonstration and writes the corresponding row directly to action. It does not replay a learned policy, regenerate actions, interpolate action values, or relabel successful trajectories. Episode boundaries are retained. Output timestamps are normalized to the declared 20 FPS.

This low-dimensional source contains no aligned image, depth, audio, or video observations; none are claimed by this release.

Validation evidence

The complete machine-readable evidence is in validation-report.json. The release passed all seven critical automated checks:

• Dataset metadata counts
• Parquet schema and frame count
• Episode boundaries and timestamp regularity
• Feature dimensions and finite values
• LeRobot v3 relational metadata
• License and provenance completeness
• Official LeRobot reader smoke test

The readiness score is 84/100 — Review recommended. Two medium-confidence statistical findings are retained rather than hidden:

• observation.state: 55.18% of frames contain at least one dimension more than six median absolute deviations from its median.
• action: 6.03% of frames contain at least one dimension more than six median absolute deviations from its median.

These robust-outlier signals are not proof of corruption. Review the per-feature distributions in the validation report before using the dataset for training.

Source, revision, and integrity

• Source dataset: robomimic/robomimic_datasets
• Pinned source revision: 74fa018461f479cd9fd15b924a16103012096203
• Source file: v1.5/lift/ph/low_dim_v15.hdf5
• Source file size: 21,084,088 bytes
• Source SHA-256: 2067777cb8b532e9263dd09fd6448c41cc31224bb27be4a3b734010ae13eb540
• Converter adapter: robomimic adapter 1.0.0
• Converter code revision: 4230bc4b56abe85116b202398b1a625a81c1c55c

Additional audit files:

• provenance.json — source revision, citation, input checksum, converter, and adapter identity
• bundle-manifest.json — generated artifact inventory and embedded validation evidence
• bundle-manifest.external.json — externally written conversion manifest
• UPSTREAM_LICENSE.md — revision-specific upstream license and attribution record

License and attribution

The pinned upstream dataset card declares the dataset as MIT. This conversion retains that identifier and the upstream citation. Conversion does not transfer ownership, create affiliation, or replace the upstream terms. Users remain responsible for reviewing the source dataset and any separately licensed simulator assets used in their own workflows.

Intended use

• Validating LeRobot v3 readers and data pipelines
• Small offline imitation-learning or educational experiments
• Reproducible testing of HDF5-to-LeRobot conversion
• Comparing schema, provenance, and quality-validation tooling

Limitations

• This is a community conversion, not an official upstream release.
• It contains only the first 10 episodes from the pinned upstream file, not the complete Lift PH dataset.
• It is a low-dimensional state-action dataset and is not multimodal.
• Statistical outlier findings require domain-aware review before training.
• Automated validation does not measure downstream policy performance, task success, or simulator replay success.
• This release preserves source actions but does not independently certify the behavioral quality of the upstream demonstrations.

Load with LeRobot

from lerobot.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("ViaCatalyst/robomimic-lift-ph-lerobot-v3")
print(dataset.meta.total_episodes, dataset.meta.total_frames)

Conversion tooling

Converted and validated with the ViaCatalyst BYOD Processing Platform, a free web workflow for converting robotics datasets to LeRobot format. For high-volume datasets, contact ViaCatalyst support through the platform.

Citation

Please cite the original robomimic work:

@inproceedings{robomimic2021,
  title={What Matters in Learning from Offline Human Demonstrations for Robot Manipulation},
  author={Ajay Mandlekar and Danfei Xu and Josiah Wong and Soroush Nasiriany and Chen Wang and Rohun Kulkarni and Li Fei-Fei and Silvio Savarese and Yuke Zhu and Roberto Mart\'in-Mart\'in},
  booktitle={Conference on Robot Learning (CoRL)},
  year={2021}
}