lerobot/pusht converted to the Lance layout that LeRobot's LanceDBDataset reads natively. Same meta/ sidecar as the original, tabular features in frames.lance, encoded videos in a videos.lance blob v2 column.
from lerobot.datasets import LeRobotDataset # not needed, shown for contrast
from lerobot.datasets import LanceDBDataset
ds = LanceDBDataset("lance-format/pusht-lance") # streams tables from the Hub, only meta/ is downloaded
item = ds[0]
Or just… See the full description on the dataset page:
https://huggingface.co/datasets/lance-format/pusht-lance.
