PointWorld-DROID WebDataset shards

WebDataset (WDS) train/test shards for training and evaluating PointWorld, converted from the nvidia/PointWorld-DROID release. This is not an official NVIDIA release.

License

NVIDIA License, non-commercial: research and evaluation use only. See LICENSE.pdf.

This data is a derivative work of nvidia/PointWorld-DROID, which is distributed under the NVIDIA License, and is redistributed under that same license. The NVIDIA License permits use of this data and of anything derived from it only for non-commercial purposes, meaning research or evaluation. The underlying robot episodes come from the DROID dataset; see its terms as well.

Contents

droid/wds/
├── train/train-rank{R}-NNNNNN.tar
├── test/test-rank{R}-NNNNNN.tar
├── test/expert_confidence-seed=42.h5      # for confidence-filtered DROID evaluation
├── metadata_rank{R}.json
└── {train,test}_source_paths_rank{R}.txt
_jobs/
├── wds_index.json                         # samples per tar
├── integrity_check.json                   # merged integrity results (valid clips)
└── ...                                    # build bookkeeping

The layout matches what the PointWorld main branch expects: point LOCAL_DATASET_DIR at a directory containing droid/wds. You can mount the bucket there or sync it locally (about 4.7 TB).

How it was made

Built with the data branch pipeline (data_integrity_check.py + convert_wds.py), run per source shard on Hugging Face Jobs. Very minor differences from the upstream release and instructions:

• Split: the paper split manifests/droid_paper_split_manifest.json. All 51,040 manifest test clips are present.
• Clip ids: released H5 files are named <episode>_flows.h5, but the paper manifest, the confidence file and main use <episode>. The _flows suffix was stripped before conversion, so WDS keys are <episode>-<start>:<end>.
• Content: WDS samples contain the fields listed by get_wds_data_keys("droid") in shared/data_contract.py. Other H5 fields (e.g. joint velocities and torques) are not included. Depth packages and camera JSONs from the source release are not included.

Citation

If you use this data, cite the PointWorld paper and the original DROID dataset (https://droid-dataset.github.io/).

@article{huang2026pointworld,
  title={PointWorld: Scaling 3D World Models for In-The-Wild Robotic Manipulation},
  author={Huang, Wenlong and Chao, Yu-Wei and Mousavian, Arsalan and Liu, Ming-Yu and Fox, Dieter and Mo, Kaichun and Li, Fei-Fei},
  journal={arXiv preprint arXiv:2601.03782},
  year={2026}
}