Add distance_to_camera and distance_to_image_plane camera presets to cartpole, dexsuite_kuka, and shadow_hand with test coverage

CONTRIBUTORS.md

@@ -159,6 +159,7 @@ Guidelines for modifications:
 * Rafael Wiltz
 * Renaud Poncelet
 * René Zurbrügg
+* Richard Lei
 * Richard Schmitt
 * RinZ27
 * Ritvik Singh

source/isaaclab_tasks/changelog.d/extend-test-distance-aovs.rst

@@ -0,0 +1,10 @@
+Added
+^^^^^
+
+* Added ``distance_to_camera`` and ``distance_to_image_plane`` camera presets to
+  :class:`~isaaclab_tasks.core.cartpole.cartpole_direct_camera_env_cfg.CartpoleTiledCameraCfg`,
+  :class:`~isaaclab_tasks.core.cartpole.cartpole_manager_camera_env_cfg.CartpoleTiledCameraCfg`,
+  :class:`~isaaclab_tasks.core.dexsuite.config.kuka_allegro.camera_cfg.BaseTiledCameraCfg`,
+  :class:`~isaaclab_tasks.core.dexsuite.config.kuka_allegro.camera_cfg.WristTiledCameraCfg`, and
+  :class:`~isaaclab_tasks.core.shadow_hand.shadow_hand_vision_env_cfg.ShadowHandVisionTiledCameraCfg`,
+  enabling golden-image rendering tests for these depth-like AOV types.

source/isaaclab_tasks/isaaclab_tasks/core/cartpole/cartpole_direct_camera_env_cfg.py

@@ -34,6 +34,8 @@ class BaseCartpoleTiledCameraCfg(CameraCfg):
 
     default = BaseCartpoleTiledCameraCfg(data_types=["rgb"])
     depth = BaseCartpoleTiledCameraCfg(data_types=["depth"])
+    distance_to_camera = BaseCartpoleTiledCameraCfg(data_types=["distance_to_camera"])
+    distance_to_image_plane = BaseCartpoleTiledCameraCfg(data_types=["distance_to_image_plane"])
     albedo = BaseCartpoleTiledCameraCfg(data_types=["albedo"])
     semantic_segmentation = BaseCartpoleTiledCameraCfg(data_types=["semantic_segmentation"])
     simple_shading_constant_diffuse = BaseCartpoleTiledCameraCfg(data_types=["simple_shading_constant_diffuse"])
@@ -73,6 +75,8 @@ class BaseCartpoleCameraEnvCfg(CartpoleEnvCfg):
 
     default = BaseCartpoleCameraEnvCfg()
     depth = BaseCartpoleCameraEnvCfg(observation_space=[1, 100, 100])
+    distance_to_camera = BaseCartpoleCameraEnvCfg(observation_space=[1, 100, 100])
+    distance_to_image_plane = BaseCartpoleCameraEnvCfg(observation_space=[1, 100, 100])
     albedo = BaseCartpoleCameraEnvCfg()
     semantic_segmentation = BaseCartpoleCameraEnvCfg(observation_space=[4, 100, 100])
     simple_shading_constant_diffuse = BaseCartpoleCameraEnvCfg()

source/isaaclab_tasks/isaaclab_tasks/core/cartpole/cartpole_manager_camera_env_cfg.py

@@ -45,6 +45,8 @@ class BaseCartpoleTiledCameraCfg(CameraCfg):
 
     default = BaseCartpoleTiledCameraCfg(data_types=["rgb"])
     depth = BaseCartpoleTiledCameraCfg(data_types=["depth"])
+    distance_to_camera = BaseCartpoleTiledCameraCfg(data_types=["distance_to_camera"])
+    distance_to_image_plane = BaseCartpoleTiledCameraCfg(data_types=["distance_to_image_plane"])
     albedo = BaseCartpoleTiledCameraCfg(data_types=["albedo"])
     semantic_segmentation = BaseCartpoleTiledCameraCfg(data_types=["semantic_segmentation"])
     simple_shading_constant_diffuse = BaseCartpoleTiledCameraCfg(data_types=["simple_shading_constant_diffuse"])
@@ -167,6 +169,8 @@ def __post_init__(self):
 
     rgb = BaseCartpoleCameraEnvCfg(observations=image_observations_cfg("rgb"))
     depth = BaseCartpoleCameraEnvCfg(observations=image_observations_cfg("depth"))
+    distance_to_camera = BaseCartpoleCameraEnvCfg(observations=image_observations_cfg("distance_to_camera"))
+    distance_to_image_plane = BaseCartpoleCameraEnvCfg(observations=image_observations_cfg("distance_to_image_plane"))
     albedo = BaseCartpoleCameraEnvCfg(observations=image_observations_cfg("albedo"))
     semantic_segmentation = BaseCartpoleCameraEnvCfg(observations=image_observations_cfg("semantic_segmentation"))
     simple_shading_constant_diffuse = BaseCartpoleCameraEnvCfg(

source/isaaclab_tasks/isaaclab_tasks/core/dexsuite/config/kuka_allegro/camera_cfg.py

@@ -109,6 +109,12 @@ class BaseTiledCameraCfg(PresetCfg):
     depth64 = BASE_CAMERA_CFG.replace(data_types=["depth"], width=64, height=64)
     depth128 = BASE_CAMERA_CFG.replace(data_types=["depth"], width=128, height=128)
     depth256 = BASE_CAMERA_CFG.replace(data_types=["depth"], width=256, height=256)
+    distance_to_camera64 = BASE_CAMERA_CFG.replace(data_types=["distance_to_camera"], width=64, height=64)
+    distance_to_camera128 = BASE_CAMERA_CFG.replace(data_types=["distance_to_camera"], width=128, height=128)
+    distance_to_camera256 = BASE_CAMERA_CFG.replace(data_types=["distance_to_camera"], width=256, height=256)
+    distance_to_image_plane64 = BASE_CAMERA_CFG.replace(data_types=["distance_to_image_plane"], width=64, height=64)
+    distance_to_image_plane128 = BASE_CAMERA_CFG.replace(data_types=["distance_to_image_plane"], width=128, height=128)
+    distance_to_image_plane256 = BASE_CAMERA_CFG.replace(data_types=["distance_to_image_plane"], width=256, height=256)
     albedo64 = BASE_CAMERA_CFG.replace(data_types=["albedo"], width=64, height=64)
     albedo128 = BASE_CAMERA_CFG.replace(data_types=["albedo"], width=128, height=128)
     albedo256 = BASE_CAMERA_CFG.replace(data_types=["albedo"], width=256, height=256)
@@ -180,6 +186,12 @@ class WristTiledCameraCfg(PresetCfg):
     semantic_segmentation64 = WRIST_CAMERA_CFG.replace(data_types=["semantic_segmentation"], width=64, height=64)
     semantic_segmentation128 = WRIST_CAMERA_CFG.replace(data_types=["semantic_segmentation"], width=128, height=128)
     semantic_segmentation256 = WRIST_CAMERA_CFG.replace(data_types=["semantic_segmentation"], width=256, height=256)
+    distance_to_camera64 = WRIST_CAMERA_CFG.replace(data_types=["distance_to_camera"], width=64, height=64)
+    distance_to_camera128 = WRIST_CAMERA_CFG.replace(data_types=["distance_to_camera"], width=128, height=128)
+    distance_to_camera256 = WRIST_CAMERA_CFG.replace(data_types=["distance_to_camera"], width=256, height=256)
+    distance_to_image_plane64 = WRIST_CAMERA_CFG.replace(data_types=["distance_to_image_plane"], width=64, height=64)
+    distance_to_image_plane128 = WRIST_CAMERA_CFG.replace(data_types=["distance_to_image_plane"], width=128, height=128)
+    distance_to_image_plane256 = WRIST_CAMERA_CFG.replace(data_types=["distance_to_image_plane"], width=256, height=256)
     # raycaster camera presets
     raycaster_depth64 = WRIST_RAYCASTER_CAMERA_CFG.replace(pattern_cfg=RAY_PATTERN.replace(width=64, height=64))
     raycaster_depth128 = WRIST_RAYCASTER_CAMERA_CFG.replace(pattern_cfg=RAY_PATTERN.replace(width=128, height=128))

source/isaaclab_tasks/isaaclab_tasks/core/shadow_hand/shadow_hand_vision_env_cfg.py

@@ -103,6 +103,26 @@ class ShadowHandVisionTiledCameraCfg(PresetCfg):
             presets=depth,ovrtx_renderer    # depth rendering with OVRTX renderer
     """
 
+    distance_to_camera: _ShadowHandBaseTiledCameraCfg = _ShadowHandBaseTiledCameraCfg(data_types=["distance_to_camera"])
+    """Distance to camera [m] (1 channel).
+
+    .. warning::
+        This preset is intended for **benchmarking only**. The keypoint-regression CNN
+        cannot be meaningfully trained from depth alone. Use it with
+        :class:`ShadowHandVisionBenchmarkEnvCfg` (``feature_extractor.enabled=False``).
+    """
+
+    distance_to_image_plane: _ShadowHandBaseTiledCameraCfg = _ShadowHandBaseTiledCameraCfg(
+        data_types=["distance_to_image_plane"]
+    )
+    """Distance to image plane [m] (1 channel).
+
+    .. warning::
+        This preset is intended for **benchmarking only**. The keypoint-regression CNN
+        cannot be meaningfully trained from depth alone. Use it with
+        :class:`ShadowHandVisionBenchmarkEnvCfg` (``feature_extractor.enabled=False``).
+    """
+
     semantic_segmentation: _ShadowHandBaseTiledCameraCfg = _ShadowHandBaseTiledCameraCfg(
         data_types=["semantic_segmentation"]
     )
@@ -135,7 +155,10 @@ def validate_config(self):
                     "Choose a compatible preset, e.g. presets=newton_renderer,rgb."
                 )
 
-        if set(self.tiled_camera.data_types) == {"depth"} and self.feature_extractor.enabled:
+        non_depth_data_types = set(self.tiled_camera.data_types).difference(
+            {"depth", "distance_to_image_plane", "distance_to_camera"}
+        )
+        if self.tiled_camera.data_types and not non_depth_data_types and self.feature_extractor.enabled:
             raise ValueError(
                 "Depth-only camera data type is intended for benchmarking only. "
                 "The keypoint-regression CNN cannot be meaningfully trained from depth alone. "

source/isaaclab_tasks/test/rendering_test_utils.py

@@ -81,6 +81,8 @@
     "simple_shading_full_mdl",
     "semantic_segmentation",
     "depth",
+    "distance_to_camera",
+    "distance_to_image_plane",
 )
 
 
@@ -632,7 +634,7 @@ def rendering_test_shadow_hand(
     if renderer == "ovrtx_renderer":
         _redirect_ovrtx_renderer_log_to_stdout(env_cfg)
 
-    if data_type == "depth":
+    if data_type in {"depth", "distance_to_camera", "distance_to_image_plane"}:
         # Disable CNN forward pass as it cannot be meaningfully trained from depth alone and will raise a ValueError.
         env_cfg.feature_extractor.enabled = False