spikefly/SemanticVLA-Bridge-LeRobot

SemanticVLA · Bridge (LeRobot + dense traces)

🚧 Placeholder. The URL is stable; data will be uploaded incrementally per the release roadmap.

LeRobot conversion of BridgeData V2 (Open X-Embodiment slug bridge_orig), paired with dense end-effector trace annotations aligned to every (episode, step) index. Used to train the trace head and the trace-conditioned Latent Action Model (LAM) in SemanticVLA.

Field	Value
Robot	WidowX
Episodes	~53,192
Frames	~1,999,410
Cameras	front RGB (224×224 in LeRobot conversion)
Action space	7-DoF end-effector delta (Δx Δy Δz Δrx Δry Δrz gripper)
Trace alignment	per (episode, step), present for 100% of episodes

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 · Fractal	https://huggingface.co/datasets/spikefly/SemanticVLA-Fractal-LeRobot
Sibling dataset · BC-Z	https://huggingface.co/datasets/spikefly/SemanticVLA-BCZ-LeRobot
Sibling dataset · DROID	https://huggingface.co/datasets/spikefly/SemanticVLA-DROID-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-Bridge-LeRobot/
├── meta/                       # info.json, episodes.jsonl, tasks.jsonl, stats.json, modality.json
├── data/                       # per-episode parquet shards
├── videos/                     # per-episode H264 mp4s
└── traces/
    ├── bridge_coords.npy       # (T, 2) end-effector trace coords, image-space
    ├── bridge_offsets.npy      # per-episode offsets into coords
    └── bridge_present.npy      # boolean mask of episodes with traces

How to use (placeholder API)

from huggingface_hub import snapshot_download

local = snapshot_download(
    repo_id="spikefly/SemanticVLA-Bridge-LeRobot",
    repo_type="dataset",
)
# then load with lerobot.common.datasets.lerobot_dataset.LeRobotDataset

Citation

If you use this dataset, please cite both SemanticVLA and the upstream BridgeData paper:

@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{walke2023bridgedata,
  title     = {BridgeData V2: A Dataset for Robot Learning at Scale},
  author    = {Walke, Homer and Black, Kevin and others},
  booktitle = {Conference on Robot Learning (CoRL)},
  year      = {2023}
}

License

Released under the MIT License, subject to the upstream BridgeData V2 license.