spikefly/SemanticVLA-DROID-LeRobot

SemanticVLA · DROID (LeRobot + dense traces)

🚧 Placeholder. The URL is stable; data will be uploaded incrementally per the release roadmap. A partial release (prefix shards) will land first, followed by the full conversion after additional QC.

LeRobot conversion of the DROID dataset, paired with dense end-effector trace annotations. Used to train the trace head and the trace-conditioned Latent Action Model (LAM) in SemanticVLA.

Field	Value
Robot	Franka Emika Panda (7-DoF)
Episodes (full)	~92,000
Frames (full)	~28M
Cameras	2× side RGB + 1× wrist RGB (DROID original)
Action space	7-DoF joint or end-effector (matched to DROID action mode)
Trace alignment	per (episode, step), generated for the released shards

Related repositories

Resource	Link
Code	https://github.com/Fei-Ni/SemanticVLA_Offcial
All-in-one dataset (4 subsets bundled)	https://huggingface.co/datasets/spikefly/SemanticVLA-TraceX-260K
Sibling dataset · Bridge	https://huggingface.co/datasets/spikefly/SemanticVLA-Bridge-LeRobot
Sibling dataset · Fractal	https://huggingface.co/datasets/spikefly/SemanticVLA-Fractal-LeRobot
Sibling dataset · BC-Z	https://huggingface.co/datasets/spikefly/SemanticVLA-BCZ-LeRobot
Checkpoints	https://hf.co/collections/spikefly/semanticvla-model-zoo
Collection · Datasets	https://hf.co/collections/spikefly/semanticvla-datasets
Collection · Model Zoo	https://hf.co/collections/spikefly/semanticvla-model-zoo

Planned layout

SemanticVLA-DROID-LeRobot/
├── meta/                       # info.json, episodes.jsonl, tasks.jsonl, stats.json, modality.json
├── data/                       # per-episode parquet shards
├── videos/                     # per-episode H264 mp4s
└── traces/
    ├── droid_coords.npy
    ├── droid_offsets.npy
    └── droid_present.npy

How to use (placeholder API)

from huggingface_hub import snapshot_download

local = snapshot_download(
    repo_id="spikefly/SemanticVLA-DROID-LeRobot",
    repo_type="dataset",
)

Citation

@misc{semanticvla2026,
  title  = {SemanticVLA: Semantic Vision-Language-Action Models with Trace-Conditioned Latent Action Tokens},
  author = {SemanticVLA Authors},
  year   = {2026},
  url    = {https://github.com/Fei-Ni/SemanticVLA_Offcial}
}

@inproceedings{khazatsky2024droid,
  title     = {DROID: A Large-Scale In-the-Wild Robot Manipulation Dataset},
  author    = {Khazatsky, Alexander and Pertsch, Karl and others},
  booktitle = {Robotics: Science and Systems (RSS)},
  year      = {2024}
}

License

Released under the MIT License, subject to the upstream DROID license.