feat(dataset): add slice support to LeRobotDataset.__getitem__ (#4129)

* feat(dataset): add efficient slice support

* fix(dataset): handle empty dataset slices

* refactor(dataset): reuse scalar path for slices

---------

Co-authored-by: Francesco Capuano <fc.francescocapuano@gmail.com>

src/lerobot/datasets/lerobot_dataset.py

@@ -478,18 +478,19 @@ class LeRobotDataset(torch.utils.data.Dataset):
         """Return the number of frames in the selected episodes."""
         return self.num_frames
 
-    def __getitem__(self, idx) -> dict:
+    def __getitem__(self, idx: int | slice) -> dict | list[dict]:
-        """Return a single frame by index, with all transforms applied.
+        """Return one frame or a slice of frames, with all transforms applied.
 
         Loads the frame from the underlying HF dataset, expands delta-timestamp
         windows, decodes video frames, and applies image transforms. Delegates
-        the core logic to :meth:`DatasetReader.get_item`.
+        the core logic to :class:`DatasetReader`.
 
         Args:
-            idx: Index into the (possibly episode-filtered) dataset.
+            idx: Integer index or slice into the possibly episode-filtered dataset.
 
         Returns:
-            Dict mapping feature names to their tensor values for this frame.
+            A frame dictionary for an integer index, or a list of frame
+            dictionaries for a slice.
 
         Raises:
             RuntimeError: If the dataset is currently being recorded and
@@ -499,6 +500,9 @@ class LeRobotDataset(torch.utils.data.Dataset):
             raise RuntimeError(
                 "Cannot read from a dataset that is being recorded. Call finalize() first, then access items."
             )
+        if isinstance(idx, slice):
+            return [self[item_idx] for item_idx in range(*idx.indices(len(self)))]
+
         reader = self._ensure_reader()
         if reader.hf_dataset is None:
             # One-shot load after finalize()

tests/datasets/test_datasets.py

@@ -114,6 +114,20 @@ def test_dataset_initialization(tmp_path, lerobot_dataset_factory):
     assert dataset.num_frames == len(dataset)
 
 
+def test_dataset_slice(tmp_path, lerobot_dataset_factory):
+    dataset = lerobot_dataset_factory(
+        root=tmp_path / "test", total_episodes=3, total_frames=30, use_videos=False
+    )
+
+    assert len(dataset[:5]) == 5
+    assert len(dataset[::2]) == (len(dataset) + 1) // 2
+    assert [item["index"].item() for item in dataset[4::-1]] == [4, 3, 2, 1, 0]
+    assert [item["index"].item() for item in dataset[-3:]] == list(range(len(dataset) - 3, len(dataset)))
+    assert dataset[len(dataset) :] == []
+    assert isinstance(dataset[0], dict)
+    assert dataset[:1][0].keys() == dataset[0].keys()
+
+
 # TODO(rcadene, aliberts): do not run LeRobotDataset.create, instead refactor LeRobotDatasetMetadata.create
 # and test the small resulting function that validates the features
 def test_dataset_feature_with_forward_slash_raises_error():
@@ -1741,6 +1755,38 @@ def test_delta_timestamps_query_returns_correct_values(tmp_path, empty_lerobot_d
     assert is_pad == [True, False], f"Expected [True, False], got {is_pad}"
 
 
+def test_dataset_slice_with_delta_timestamps(tmp_path, empty_lerobot_dataset_factory):
+    features = {
+        "observation.state": {"dtype": "float32", "shape": (1,), "names": ["x"]},
+    }
+    dataset = empty_lerobot_dataset_factory(
+        root=tmp_path / "test_slice_delta", features=features, use_videos=False, fps=10
+    )
+
+    for frame_idx in range(5):
+        dataset.add_frame(
+            {
+                "observation.state": torch.tensor([frame_idx], dtype=torch.float32),
+                "task": "task_0",
+            }
+        )
+    dataset.save_episode()
+    dataset.finalize()
+
+    sliced_dataset = LeRobotDataset(
+        dataset.repo_id,
+        root=dataset.root,
+        delta_timestamps={"observation.state": [-0.1, 0.0]},
+        tolerance_s=0.04,
+    )
+
+    items = sliced_dataset[:2]
+
+    assert items[0]["observation.state"].tolist() == [0.0, 0.0]
+    assert items[0]["observation.state_is_pad"].tolist() == [True, False]
+    assert items[1]["observation.state"].tolist() == [0.0, 1.0]
+
+
 def test_episode_filter_filters_dataset(tmp_path, lerobot_dataset_factory):
     """episode_filter on LeRobotDataset narrows the loaded dataset to matching episodes."""
     dataset = lerobot_dataset_factory(root=tmp_path / "test", total_episodes=8, total_frames=200)