diff --git a/Mapping/point_cloud_sampling/point_cloud_sampling.py b/Mapping/point_cloud_sampling/point_cloud_sampling.py
index df7cde41c0..e126346809 100644
--- a/Mapping/point_cloud_sampling/point_cloud_sampling.py
+++ b/Mapping/point_cloud_sampling/point_cloud_sampling.py
@@ -5,6 +5,8 @@
 - Poisson disk sampling
 
 """
+import operator
+
 import matplotlib.pyplot as plt
 import numpy as np
 import numpy.typing as npt
@@ -48,16 +50,28 @@ def farthest_point_sampling(orig_points: npt.NDArray,
     ----------
     orig_points :  (M, N) N-dimensional points for sampling.
                     The number of points is M.
-    n_points : number of points for sampling
+    n_points : integer number of points for sampling, from 0 to M inclusive.
     seed : random seed number
 
     Returns
     -------
-    sampled points (n_points, N)
+    sampled points (n_points, N), with the input dtype.
+
+    Raises
+    ------
+    TypeError
+        If n_points is not an integer.
+    ValueError
+        If n_points is negative or exceeds the number of input rows.
 
     """
-    rng = np.random.default_rng(seed)
+    n_points = operator.index(n_points)
     n_orig_points = orig_points.shape[0]
+    if not 0 <= n_points <= n_orig_points:
+        raise ValueError("n_points must be between 0 and the number of input rows")
+    if n_points == 0:
+        return orig_points[:0].copy()
+    rng = np.random.default_rng(seed)
     first_point_id = rng.choice(range(n_orig_points))
     min_distances = np.ones(n_orig_points) * float("inf")
     selected_ids = [first_point_id]
diff --git a/docs/modules/3_mapping/point_cloud_sampling/point_cloud_sampling_main.rst b/docs/modules/3_mapping/point_cloud_sampling/point_cloud_sampling_main.rst
index 8cb08d4b78..f15df69519 100644
--- a/docs/modules/3_mapping/point_cloud_sampling/point_cloud_sampling_main.rst
+++ b/docs/modules/3_mapping/point_cloud_sampling/point_cloud_sampling_main.rst
@@ -44,6 +44,12 @@ possible.
 This method is useful for machine learning and other situations where
 you want to obtain a specified number of points from point cloud.
 
+The requested count must be an integer between zero and the number of input
+rows, inclusive. Each input row is selected at most once, even when multiple
+rows have identical coordinates. A zero count returns an empty array with the
+same coordinate dimension and dtype as the input. Negative or excessive counts
+raise ``ValueError`` instead of starting a sampling loop that cannot finish.
+
 API
 =====
 
diff --git a/tests/test_point_cloud_sampling.py b/tests/test_point_cloud_sampling.py
index 8f6447c69c..0a9c87925d 100644
--- a/tests/test_point_cloud_sampling.py
+++ b/tests/test_point_cloud_sampling.py
@@ -1,3 +1,6 @@
+import numpy as np
+import pytest
+
 import conftest  # Add root path to sys.path
 from Mapping.point_cloud_sampling import point_cloud_sampling as m
 
@@ -11,5 +14,47 @@ def test_1(capsys):
     assert "poisson_disk_points.shape=(20, 3)" in captured.out
 
 
+@pytest.mark.parametrize("points", [np.empty((0, 3)), np.ones((4, 2), dtype=np.int32)])
+def test_farthest_sampling_zero_count(points):
+    sampled = m.farthest_point_sampling(points, 0, seed=0)
+    assert sampled.shape == (0, points.shape[1])
+    assert sampled.dtype == points.dtype
+
+
+@pytest.mark.parametrize("count", [-1, 3])
+def test_farthest_sampling_rejects_unavailable_count(count):
+    with pytest.raises(ValueError, match="between 0 and"):
+        m.farthest_point_sampling(np.zeros((2, 3)), count, seed=0)
+
+
+def test_farthest_sampling_rejects_nonempty_sample_from_empty_cloud():
+    with pytest.raises(ValueError, match="between 0 and"):
+        m.farthest_point_sampling(np.empty((0, 3)), 1, seed=0)
+
+
+@pytest.mark.parametrize("count", [1.5, 2.0, "2", None])
+def test_farthest_sampling_requires_integer_count(count):
+    with pytest.raises(TypeError):
+        m.farthest_point_sampling(np.zeros((3, 2)), count, seed=0)
+
+
+@pytest.mark.parametrize("seed, expected", [
+    (0, [[16, 0], [0, 0], [9, 0], [4, 0]]),
+    (42, [[0, 0], [16, 0], [9, 0], [4, 0]]),
+])
+def test_farthest_sampling_preserves_seeded_order(seed, expected):
+    points = np.array([[0, 0], [1, 0], [4, 0], [9, 0], [16, 0]])
+    np.testing.assert_array_equal(
+        m.farthest_point_sampling(points, np.int64(4), seed), expected)
+
+
+def test_farthest_sampling_duplicate_points_use_distinct_rows():
+    points = np.array([[0, 0], [0, 0], [1, 0], [1, 0]])
+    original = points.copy()
+    sampled = m.farthest_point_sampling(points, len(points), seed=0)
+    np.testing.assert_array_equal(np.sort(sampled, axis=0), points)
+    np.testing.assert_array_equal(points, original)
+
+
 if __name__ == '__main__':
     conftest.run_this_test(__file__)