fix(datasets): support features with a zero-width dimension (shape=(0,))

Declaring a numeric feature with `shape=(0,)` crashed `save_episode()` in two
distinct places, leaving `dtype: "string"` as the only (type-lossy) workaround:

  - `compute_episode_stats` -> `RunningQuantileStats.update` reshaped a size-0
    array, raising "ValueError: cannot reshape array of size 0 into shape (0)".
  - `get_hf_features_from_features` mapped it to a fixed-size Arrow list of
    length 0, which pyarrow rejects ("list_size needs to be a strict positive
    integer").

The issue only reported the first error; the second surfaces once the first is
fixed. This change handles both:

  - Skip zero-width features during episode stats, exactly as string/language
    features are already skipped.
  - Store 1-D zero-width features as a variable-length sequence (length=-1) so
    each per-frame value is simply an empty list.

Adds a unit test (stats layer) and an integration test that records, saves, and
reads back a zero-width feature, asserting it round-trips as an empty vector and
is excluded from stats.

Fixes #3654

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/datasets/test_compute_stats.py

@@ -643,6 +643,31 @@ def test_compute_episode_stats_string_features_skipped():
     assert "q01" in stats["action"]
 
 
+def test_compute_episode_stats_zero_width_feature_skipped():
+    """Features with a zero-width dimension carry no values and must be skipped, not crash.
+
+    Regression test for https://github.com/huggingface/lerobot/issues/3654:
+    a feature declared with shape=(0,) previously reached RunningQuantileStats.update,
+    where ``batch.reshape(-1, batch.shape[-1])`` raised
+    "ValueError: cannot reshape array of size 0 into shape (0)".
+    """
+    episode_data = {
+        "action": np.random.normal(0, 1, (100, 5)).astype(np.float32),
+        "target": np.zeros((100, 0), dtype=np.float32),  # zero-width feature
+    }
+    features = {
+        "action": {"dtype": "float32", "shape": (5,)},
+        "target": {"dtype": "float32", "shape": (0,)},
+    }
+
+    stats = compute_episode_stats(episode_data, features)
+
+    # Zero-width features are skipped, just like strings; non-empty features are unaffected.
+    assert "target" not in stats
+    assert "action" in stats
+    assert "q01" in stats["action"]
+
+
 def test_aggregate_feature_stats_with_quantiles():
     """Test aggregating feature stats that include quantiles."""
     stats_ft_list = [

tests/datasets/test_dataset_writer.py

@@ -189,6 +189,34 @@ def test_save_multiple_episodes(tmp_path):
     assert dataset.meta.total_frames == total_frames
 
 
+def test_save_episode_with_zero_width_feature(tmp_path):
+    """save_episode() succeeds when a feature has a zero-width dimension (shape=(0,)).
+
+    Regression test for https://github.com/huggingface/lerobot/issues/3654: such a
+    feature previously crashed stats computation with
+    "ValueError: cannot reshape array of size 0 into shape (0)".
+    """
+    features = {
+        **SIMPLE_FEATURES,
+        "target": {"dtype": "float32", "shape": (0,), "names": None},
+    }
+    root = tmp_path / "ds"
+    dataset = LeRobotDataset.create(repo_id=DUMMY_REPO_ID, fps=DEFAULT_FPS, features=features, root=root)
+    for _ in range(4):
+        dataset.add_frame(_make_frame(features))
+    dataset.save_episode()  # previously raised ValueError on the zero-width 'target' feature
+    dataset.finalize()
+
+    assert dataset.meta.total_episodes == 1
+    assert dataset.meta.total_frames == 4
+
+    # The zero-width feature round-trips back as an empty vector and is excluded from stats.
+    reloaded = LeRobotDataset(repo_id=DUMMY_REPO_ID, root=root)
+    target = np.asarray(reloaded[0]["target"])
+    assert target.shape == (0,)
+    assert "target" not in (reloaded.meta.stats or {})
+
+
 # ── clear / lifecycle ────────────────────────────────────────────────