[python] Add Paimon LeRobot map-style dataset

docs/docs/pypaimon/multimodal-api.mdx

@@ -562,6 +562,28 @@ keep their compressed bytes.
 Only v3 is supported. Video features, `uint64`, and language event structures
 are rejected.
 
+For map-style training, reuse the source dataset's `meta/` contract while
+reading frames lazily from the imported table:
+
+```python
+from torch.utils.data import DataLoader
+from pypaimon.multimodal import PaimonLeRobotDataset
+
+dataset = PaimonLeRobotDataset(
+    conn.get_table("robot_data"),
+    "/data/lerobot_dataset/meta",
+    blob_parallelism=8,
+)
+loader = DataLoader(dataset, batch_size=32, shuffle=True, num_workers=4)
+```
+
+The metadata argument supplies the standard info, episodes, tasks, and stats
+through `dataset.meta`; the imported table does not persist the complete
+`meta/` directory. Payload columns remain lazy. Pass
+`index_mapping=dataset.index_mapping` to readers of the same table snapshot to
+avoid rescanning the control columns. The first version supports image-backed datasets;
+video-backed reads remain a follow-up.
+
 ## Overwrite
 
 `overwrite` accepts the same input formats as `add` and replaces existing data

paimon-python/pypaimon/multimodal/__init__.py

@@ -29,6 +29,7 @@
     Hdf5File,
     Hdf5LoadResult,
 )
+from pypaimon.multimodal.lerobot.dataset import PaimonLeRobotDataset
 from pypaimon.multimodal.table import (
     MultimodalTable,
     TextRoute,
@@ -55,6 +56,7 @@
     "MultimodalTable",
     "NoSuchKey",
     "ObjectInfo",
+    "PaimonLeRobotDataset",
     "PutObjectResult",
     "TextRoute",
     "VectorRoute",

paimon-python/pypaimon/multimodal/lerobot/__init__.py

@@ -14,11 +14,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""One-time LeRobot Dataset v3 import into a multimodal Paimon table."""
+"""LeRobot Dataset v3 integration for multimodal Paimon tables."""
 
 from pypaimon.multimodal.lerobot.api import load_from_lerobot
+from pypaimon.multimodal.lerobot.dataset import PaimonLeRobotDataset
 
 
 __all__ = [
+    "PaimonLeRobotDataset",
     "load_from_lerobot",
 ]